# Changelog

All notable changes to this project will be documented in this file.

## [unreleased]

## [2.2.366] - 2026-10-27
### General
#### Added
- Added distroless variants of the analyzer and anonymizer container images, built from `Dockerfile.distroless` on Azure Linux and published to the same GHCR repositories as `<version>-distroless` and `latest-distroless` tags; the unsuffixed tags, including `latest`, remain the default images. The distroless images contain no shell, package manager, pip or uv, run Python 3.12 as UID 65532, listen on port 3000, and take gunicorn settings through `GUNICORN_CMD_ARGS` instead of `PORT` and `WORKERS`. A `docker-compose.distroless.yml` override runs them locally, and CI runs the e2e tests against them (#2407) (Thanks @SharonHart)

#### Changed
- Dropped Python 3.10 support, which reached end of life on 2026-10-01. All packages now require Python `>=3.11,<3.15`; Python 3.10 users resolve to the previous releases (#2409) (Thanks @SharonHart)
- The default analyzer and anonymizer images no longer contain `curl`: the image `HEALTHCHECK` now probes `/health` with the Python standard library, ignoring any `HTTP_PROXY` settings, and the `docker-compose.yml` healthcheck overrides were removed so the image healthchecks apply (#2407) (Thanks @SharonHart)

#### Fixed
- Fixed the documentation publishing workflow deleting the `CNAME` file, which took the custom `presidio.dataprivacystack.org` domain offline after every publish; canonical links and the sitemap now use the custom domain (#2410) (Thanks @omri374)
- Fixed the documented `docker run` port mapping for a source-built anonymizer image, which listens on container port 3000 (#2407) (Thanks @SharonHart)

#### Security
- Slimmed and hardened the default analyzer and anonymizer images: the `python:3.12-slim` and `python:3.14-slim` bases are re-pinned with OS security updates applied, uv is bind-mounted at build time instead of installed, and setuptools and wheel (plus pip in the anonymizer) are removed from the final image. This cuts the Trivy critical, high and medium findings from 321 to 103 (anonymizer) and from 345 to 113 (analyzer); the distroless images bring them down to 1 and 6 (#2407) (Thanks @SharonHart)
- Refreshed every `uv.lock` file to fix the 100 known advisories in the locked development and CI dependencies, including `pillow` 13.0.0, `virtualenv` 21.14.13, `pyjwt` 2.15.1, `aiohttp` 3.14.5, `urllib3` 2.8.0, `cryptography` 50.0.2, `filelock` 4.2.4, `oauthlib` 4.0.0 and `certifi` 2026.10.23; apart from the `pillow` and `onnxruntime` requirements listed below, published dependency ranges are unchanged (#2409) (Thanks @SharonHart)

### Analyzer
#### Added
- Added an optional `context_aware_enhancer` section to the analyzer YAML configuration so the `LemmaContextAwareEnhancer` parameters (`context_similarity_factor`, `min_score_with_context_similarity`, `context_prefix_count`, `context_suffix_count` and `context_matching_mode`) can be set without Python code; invalid values are rejected at parse time with an actionable message, and configurations without the section keep the previous defaults (#2373) (Thanks @MetehanAydemir)

#### Changed
- Replaced the unmaintained `spacy-huggingface-pipelines` dependency with an internal spaCy token-classification component. `TransformersNlpEngine` now registers the `presidio_hf_token_pipe` factory instead of `hf_token_pipe`, so a spaCy pipeline saved with `nlp.to_disk()` that references `hf_token_pipe` no longer loads. The `transformers` extra now requires `transformers>=4.28.0,<6.0.0`, and the `gliner` extra requires `gliner>=0.2.29` and `transformers>=4.51.3,<5.17.0`. Detection results and scores are unchanged (#2201) (Thanks @SharonHart)
- The `gliner` extra now declares a single `onnxruntime>=1.19` requirement; the `<1.24.1` cap applied only to Python 3.10 (#2409) (Thanks @SharonHart)

#### Security
- Locked `transformers` to 5.16.1, which fixes GHSA-xrqw-3rrv-vx5w and GHSA-x9r9-c232-4q39; the published range still allows Transformers 4.x, and installing Transformers 5.10 or later is recommended to pick up the fixes (#2201) (Thanks @SharonHart)

### Image Redactor
#### Security
- Raised the `Pillow` requirement to `>=13.0.0,<14.0.0`, which fixes the 41 security advisories (5 critical, 23 high) open against Pillow 12.3.0, and replaced the `Image.getdata()` call in the DICOM redactor with `get_flattened_data()`, since `getdata()` is deprecated in Pillow 13 and removed in Pillow 14 (#2393, #2409) (Thanks @SharonHart)

## [2.2.365] - 2026-10-25
### General
#### Added
- Added a documentation sample showing how to use Presidio Analyzer and Anonymizer as an input guardrail for the AISIX AI Gateway (#2257) (Thanks @Yilialinn)

#### Changed
- Restructured the repository's AI guidance into Copilot review instructions and cross-agent guidelines (`AGENTS.md`, `CLAUDE.md`) covering recognizers, YAML configuration, and general contribution practices (#2211) (Thanks @omri374)
- Added automatic pull request labelling by component and area based on the changed paths (#2240) (Thanks @omri374)
- Updated the README and project transition documentation to reflect that the transition to the Data Privacy Stack is complete (#2274) (Thanks @omri374)
- Updated the development setup documentation from Poetry to uv to match what CI runs (#2220) (Thanks @juno-junho)

#### Fixed
- Fixed four `presidio-cli` bugs: `--no-warnings` crashing, the exit code not reflecting the number of problems found, `-f github` not emitting `::warning`/`::error` workflow commands, and out-of-range or non-numeric `threshold` config values being accepted silently (#2266) (Thanks @v0ropaev)

#### Security
- Updated the locked `aiohttp` to 3.14.3 and `anyio` to 4.14.2 in the `presidio-cli` lockfile to fix six vulnerabilities; published dependency ranges are unchanged (#2304) (Thanks @RonShakutai)

### Analyzer
#### Added
- Added `UuidRecognizer` (generic, entity type `UUID`) to detect UUIDs in the standard 8-4-4-4-12 hyphenated hexadecimal format, covering RFC 4122 versions 1-5 and RFC 9562 versions 6-8. Validates version and variant nibbles and filters the nil UUID to reduce false positives.
- South African ID number (`ZA_ID_NUMBER`) recognizer for the 13-digit national identity number, using pattern matching, context words, birth-date validation, and Luhn checksum validation. Disabled by default.
- South African recognizers for `ZA_PASSPORT`, `ZA_INCOME_TAX_NUMBER`, `ZA_DRIVER_LICENSE`, `ZA_VAT_NUMBER`, `ZA_COMPANY_REGISTRATION`, `ZA_TRAFFIC_REGISTER_NUMBER`, `ZA_LICENSE_PLATE`, `ZA_MOBILE_NUMBER`, and `ZA_TELEPHONE_NUMBER`. All disabled by default.
- Philippine passport (`PH_PASSPORT`) recognizer with pattern matching and context support. Disabled by default.
- Added `TokenizerBasedTextChunker` for token-accurate text chunking in NER recognizers, YAML-configurable `text_chunker` for `GLiNERRecognizer` and `HuggingFaceNerRecognizer`, and `chunk_size`/`chunk_overlap` parameters on `GLiNERRecognizer` (#2041) (Thanks @yuriihavrylko)
- Added `CaPostalCodeRecognizer` (`CA_POSTAL_CODE`) for Canadian postal codes with English and French context words; disabled by default (#2204) (Thanks @watsos)
- Added six US healthcare identifier recognizers, `US_HEALTH_INSURANCE_MEMBER_ID`, `US_PRIOR_AUTHORIZATION_NUMBER`, `US_CLAIM_NUMBER`, `US_PRESCRIPTION_NUMBER`, `US_REFERRAL_NUMBER`, and `US_PROVIDER_TAX_ID`, that rely on context words and entity-specific score thresholds; all are disabled by default (#2159) (Thanks @bhargavikalicheti)

#### Changed
- The `kr` language code is deprecated in favor of the ISO 639-1 code `ko`. `KrRrnRecognizer` and `KrPassportRecognizer` still accept `kr` in `supported_languages` as a backward-compatibility alias for registries configured against their original class defaults; the alias will be removed in <release TBD>. `KrBrnRecognizer`, `KrDriverLicenseRecognizer` and `KrFrnRecognizer` have only ever defaulted to `ko`, so `kr` was removed from their `default_recognizers.yaml` entries in this release. (#2236)
- Requesting a mix of supported and unsupported entities still returns partial results, but `RecognizerRegistry.get_recognizers` now logs a warning naming the unsupported entity and language, noting that ignoring unsupported entities is deprecated and will raise in a future version (#2259) (Thanks @nikolas-sapa)
- Documented how to select a custom recognizer registry in the analyzer server with the `RECOGNIZER_REGISTRY_CONF_FILE` environment variable (#2311) (Thanks @nikolas-sapa)
- Documented the `global_regex_flags` and `deny_list_score` request fields for ad-hoc `PatternRecognizer` entries in the analyzer API docs (#2255) (Thanks @MohammedAlkindi)

#### Fixed
- Fixed `KrPassportRecognizer`, `KrBrnRecognizer`, `KrDriverLicenseRecognizer`, and `UsMbiRecognizer` raising a `TypeError` when built from a registry configuration by accepting the loader's `name` keyword, and registered `KrPassportRecognizer` in `default_recognizers.yaml` (#2170) (Thanks @developer0hye)
- Fixed `AzureAILanguageRecognizer` raising a `TypeError` when loaded from a registry configuration (#2224) (Thanks @developer0hye)
- Registered `AbaRoutingRecognizer`, `FiPersonalIdentityCodeRecognizer`, and `SgUenRecognizer` in `default_recognizers.yaml` (disabled by default) so they can be enabled from YAML configuration (#2218) (Thanks @juno-junho)
- Fixed the `UK_NINO` pattern matching a leading space, which shifted the reported span and corrupted anonymized output (#2209) (Thanks @inesaranab)
- Fixed `CharacterBasedTextChunker` starting overlapping chunks in the middle of a word by snapping the chunk start to a word boundary (#2226) (Thanks @inesaranab)
- Fixed an omitted `deny_list_score` in a YAML or dict recognizer registry being passed as `0.0`, which made deny-list matches score below any positive threshold; it now defaults to `1.0`, matching `PatternRecognizer` (#2254) (Thanks @ppradyoth)
- Fixed quadratic duplicate-result checks in `EntityRecognizer.remove_duplicates` by limiting containment checks to overlapping results of the same entity type; results tied on score, start, and span length are now ordered deterministically by entity type (#2280) (Thanks @mbukowicz)
- Fixed chunked NER recognizers (`GLiNERRecognizer`, `HuggingFaceNerRecognizer`) dropping the full-span prediction of an entity that straddles a chunk boundary in favor of a higher-scoring partial prediction, which left part of the entity in clear (#2362) (Thanks @Support-itecz)
- Pinned `preshed` below 3.0.14 in `presidio-analyzer`, because 3.0.14 breaks spaCy 3.8.x model loading with an `OverflowError` in `Tokenizer._flush_specials` for fresh `pip install` environments; the pin can be removed once spaCy or preshed ships a fix (#2405) (Thanks @omri374)

#### Security
- Updated the locked `aiohttp` to 3.14.3 and `anyio` to 4.14.2 in the `presidio-analyzer` lockfile to fix six vulnerabilities; published dependency ranges are unchanged (#2304) (Thanks @RonShakutai)

### Anonymizer
#### Added
- Added an optional `score` field to `OperatorResult`, carried over from the originating `RecognizerResult` and included in `to_dict()` and `from_json()` (#2059) (Thanks @kshvg)

#### Fixed
- Fixed `ConflictResolutionStrategy.REMOVE_INTERSECTIONS` trimming overlapping results to zero length and leaving the flagged text in clear by sorting on both start and end (#2331) (Thanks @anishmehta24)
- Fixed quadratic conflict resolution in `AnonymizerEngine` by resolving overlapping results with sorted sweeps, so large result sets no longer take tens of seconds (#2371) (Thanks @HardMax71)

#### Security
- Raised the `cryptography` constraint to `>=50.0.0,<51.0.0` to resolve GHSA-g6cj-pr64-35w5, which is only patched in 50.0.0 (#2231) (Thanks @Aryan-Pardeshi)

### Image Redactor
#### Security
- Raised the minimum `Pillow` version to 12.3.0, which patches 13 open security advisories, and canonicalized the requirement syntax (#2188, #2193) (Thanks @SharonHart)
- Updated the locked `aiohttp` to 3.14.3 and `anyio` to 4.14.2 in the `presidio-image-redactor` lockfile to fix six vulnerabilities; published dependency ranges are unchanged (#2304) (Thanks @RonShakutai)

### Presidio-Structured
#### Added
- Added an optional `encoding` argument to `CsvReader` and `JsonReader` that is used for every file the reader reads, while an explicit `encoding=` passed to `read()` still takes precedence (#2306) (Thanks @v0ropaev)

#### Fixed
- Fixed `PandasDataProcessor` raising an `AttributeError` for columns whose names are not valid Python identifiers (e.g. "Full Name" or "e-mail") by reading cells by label instead of through `itertuples` attributes (#2142) (Thanks @uwezkhan)

#### Security
- Updated the locked `aiohttp` to 3.14.3 and `anyio` to 4.14.2 in the `presidio-structured` lockfile to fix six vulnerabilities; published dependency ranges are unchanged (#2304) (Thanks @RonShakutai)

## [2.2.364] - 2026-07-22
### General
#### Added
- Added a `--threshold` flag to `presidio-cli` to override the analyzer confidence threshold directly from the command line (#2114) (Thanks @rodboev)

#### Changed
- Migrated CI and service-image dependency installation from Poetry to [uv](https://github.com/astral-sh/uv) with committed lockfiles, fixing prolonged dependency-resolution hangs and making builds reproducible.
- Added Python 3.14 package support for `presidio-anonymizer`, `presidio-image-redactor`, `presidio-cli`, `presidio-structured`, and `presidio` by allowing Python `<3.15` and excluding `spacy==3.8.14` on Python 3.14 where applicable (#2096) (Thanks @Copilot)
- Added `healthcheck` and `restart: unless-stopped` policies to all services in `docker-compose.yml` so containers auto-recover on crash and report readiness (#2149) (Thanks @hiro-nikaitou)
- Updated the `LICENSE` file copyright from "Microsoft Corporation" to "Presidio Contributors" (#2134) (Thanks @cobypeled)
- Replaced `presidio@microsoft.com` author email with `presidio@dataprivacystack.org` in all package `pyproject.toml` files (#2128) (Thanks @Copilot)
- Improved CI stability by optimizing disk usage, pinning the Ollama Docker image to a specific version, and running Ollama end-to-end tests on arm64 runners (#2167) (Thanks @SharonHart)
- Updated mobile homepage layout with a compact mascot for smaller screens (#2130) (Thanks @SharonHart)

#### Fixed
- Retried the Zensical documentation build on transient crashes (e.g. SIGKILL/exit 247) so the docs release pipeline no longer fails intermittently (Thanks @Copilot)
- Fixed NLP engine initialization in the GLiNER usage sample (#2135) (Thanks @matteobachini-vianova)
- Quoted `pip install` extras examples in documentation to prevent shell expansion of bracket expressions (#2093) (Thanks @nyxst4ck)

### Analyzer
#### Added
- Added `NoOpNlpEngine` for configurations that do not require NLP engine artifacts, enabling standalone recognizers such as `HuggingFaceNerRecognizer` to run without a spaCy or Stanza model (#2071) (Thanks @ultramancode)
- Added per-recognizer and per-entity score threshold configuration in the recognizer registry YAML, with the analyzer's global `default_score_threshold` as the fallback (#2116) (Thanks @rodboev)
- Added `PhUmidRecognizer` for Philippine Unified Multi-Purpose ID (UMID/CRN) numbers in dashed and plain 12-digit formats; disabled by default (#2045) (Thanks @Surya-5555)

#### Fixed
- `PhoneRecognizer.DEFAULT_SUPPORTED_REGIONS` used `"UK"`, which is not a valid `phonenumbers` (libphonenumber) region code — region codes are ISO 3166-1 alpha-2, where the United Kingdom is `"GB"`. The `"UK"` entry was a no-op, so UK numbers in national/local format (e.g. `020 7946 0958`) were never detected by default; only international-format `+44 …` numbers matched, because they carry the country code and match under any region. Replaced `"UK"` with `"GB"`.
- Language model recognizers (`BasicLangExtractRecognizer`, `AzureOpenAILangExtractRecognizer`) configured in a recognizer registry YAML now honour `config_path` (and other recognizer-specific kwargs). Previously these entries were validated by the strict `PredefinedRecognizerConfig` schema, which has no `config_path` field and does not allow extra keys, so `config_path` was silently dropped and the recognizer fell back to its bundled default model configuration. Added a `LangExtractRecognizerConfig` model (`extra="allow"`) and registered both recognizer class names in `CONFIG_MODEL_MAP`.
- `BasicLangExtractRecognizer` now honours values under `langextract.model.provider.language_model_params` (including `timeout` and `num_ctx`). Previously these were silently dropped because `langextract.extract()` ignores its `language_model_params` argument when a pre-built `ModelConfig` is passed via `config=`, causing Ollama-backed recognizers to fall back to langextract's 120s default regardless of the configured timeout. The recognizer now merges `language_model_params` into `ModelConfig.provider_kwargs`, which is the path that reaches the provider constructor. Explicit entries under `provider.kwargs:` still take precedence. Also fixed a `TypeError` when `kwargs:` or `language_model_params:` is `null` in the YAML. (#1943, Thanks @lsternlicht)
- Fixed `UsSsnRecognizer` over-blocking valid SSNs in the `987654320`-`987654329` range due to an 8-digit instead of full 9-digit prefix in the sample-SSN denylist check (#2074) (Thanks @AUTHENSOR)
- Fixed `FiHetuRecognizer` and `SgUenRecognizer` dropping valid lower-case identifiers by normalizing the check character to upper case before checksum comparison (#2141) (Thanks @uwezkhan)
- Fixed `FiHetuRecognizer` incorrectly accepting 29 Feb on non-leap century years by deriving the full 4-digit year from the century-separator character before date validation (#2111) (Thanks @jichaowang02-lang)
- Fixed `EsNieRecognizer.validate_result` always passing the digit guard because `isdigit` was referenced instead of called, raising a `TypeError` for invalid inputs instead of returning `False` (#2146) (Thanks @MohamedAklamaash)
- Fixed `DateRecognizer` ISO 8601 pattern accepting impossible month/day values such as month 00, 13-19 or day 00, 32-39 (#2113) (Thanks @jichaowang02-lang)
- Fixed `EsNifRecognizer` and `EsNieRecognizer` dropping valid lowercase Spanish identifiers by upper-casing the check letter before the mod-23 checksum (#2076) (Thanks @AUTHENSOR)
- Fixed `AuAbnRecognizer` checksum incorrectly remapping a leading-zero first digit to 9 instead of subtracting 1 per the official ABR algorithm (#2109) (Thanks @jichaowang02-lang)
- Fixed `NgNinRecognizer` rejecting valid NIINs with a leading zero caused by `int()` conversion stripping the leading zero before Verhoeff checksum validation (#2106) (Thanks @jichaowang02-lang)
- Fixed Thai TNIN pattern rejecting valid province codes 22, 52, and 58 due to over-narrow character-class ranges (#2107) (Thanks @jichaowang02-lang)
- Fixed `FiHetuRecognizer` pattern accepting illegal century separators because `+-A` in the character class was interpreted as a range rather than three separate literals (#2108) (Thanks @jichaowang02-lang)
- Fixed `InVehicleRegistrationRecognizer` under-scoring valid registrations with zero-padded district codes by normalizing to a two-digit string before the district map lookup (#2110) (Thanks @jichaowang02-lang)
- Fixed `UkNinoRecognizer` pattern matching a numeric suffix character, producing false positives such as `AB 12 34 56 1`, caused by `{1}` quantifier placed inside the character class (#2112) (Thanks @jichaowang02-lang)
- Fixed `DeFuehrerscheinRecognizer` docstring example that contradicted the recognizer's own regex and tests (#2138) (Thanks @jichaowang02-lang)
- Fixed documentation incorrectly stating that `IpRecognizer` validates a checksum (#2133) (Thanks @Coshea46)

### Anonymizer
#### Added
- Added `BatchDeanonymizeEngine` to complement `BatchAnonymizerEngine` for batch deanonymization over lists and nested dictionaries.

#### Security
- Bumped `cryptography` lower bound to `>=48.0.1` to resolve GHSA-537c-gmf6-5ccf (HIGH, vulnerable OpenSSL statically linked into wheels below 48.0.1) (#2144) (Thanks @Copilot)

### Presidio-Structured
#### Security
- Raised the minimum supported `presidio-anonymizer` version in `presidio-structured` to `2.2.364`, so installs resolve the patched anonymizer release (#2184)

## [2.2.363] - 2026-06-28
### General
#### Added
- Optional `countries` filter on `RecognizerRegistry.load_predefined_recognizers()` to scope predefined country-specific recognizers to a subset of locales (e.g. `countries=["us", "uk"]`). The same filter is also exposed as a top-level `supported_countries` field in the recognizer-registry YAML, mirroring `supported_languages`, and as an advisory per-recognizer `country_code:` field on every predefined country-specific entry in `default_recognizers.yaml` (cross-checked against the class attribute at load time). Country tagging works via two reconciled paths: the class-level `EntityRecognizer.COUNTRY_CODE` ClassVar (canonical for predefined recognizers) and the new `country_code` constructor kwarg on `EntityRecognizer` / `PatternRecognizer` (the path for custom recognizers without a subclass — flows through `PatternRecognizer.from_dict` so YAML `type: custom` entries can declare `country_code:` directly). Conflicting values raise `ValueError` at construction time so a predefined country recognizer can never be silently re-tagged. The resolved tag is read via the `country_code()` and `is_country_specific()` instance methods, and serialized through `to_dict()` / `from_dict()` for round-tripping. Inputs to the `countries` filter are validated up front (rejects bare strings, non-iterables, non-string elements, and blank codes). Locale-agnostic recognizers and untagged custom recognizers are always loaded regardless of the filter, preserving backwards compatibility. Adds `RecognizerRegistry.get_country_codes()` for introspection and a `WARNING` log when a requested country has no matching recognizer. See `docs/analyzer/filtering_by_country.md`. Fixes #1328.
- Published source distributions alongside wheels in the PyPI release pipeline (#1924) (Thanks @Copilot)

#### Changed
- Rebranded repository documentation and links for Data Privacy Stack, added project transition notices about Presidio's move to its new home at https://presidio.dataprivacystack.org/project_transition/, migrated the docs site to Material for MkDocs/Zensical, and updated docs publishing and search verification workflows (#2097, #2098, #2100, #2102, #2104, #2117, #2118, #2119, #2120, #2121, #2122) (Thanks @omri374, @SharonHart)
- Moved container image publishing from Microsoft Container Registry to GitHub Container Registry under `ghcr.io/data-privacy-stack`, updated Docker usage documentation, and fixed GHCR authentication and provenance publishing in CI (#2103, #2123, #2124) (Thanks @SharonHart, @Copilot)
- Updated Dependabot coverage and grouping for package, Docker, Docker Compose, and GitHub Actions dependencies, with defensive dependency range controls (#1928, #1929, #1965, #1984, #2005) (Thanks @dependabot, @Copilot, @SharonHart)
- Updated GitHub Actions dependencies, Docker base images, and sample dependency pins across CI/release workflows (#1913, #1914, #1915, #1920, #1927, #1935, #1936, #1937, #1938, #1966, #1967, #1968, #1975, #1976, #1985, #1991, #1996) (Thanks @dependabot)
- Updated Copilot development instructions for the repository (#1866) (Thanks @omri374)

#### Fixed
- Skipped the coverage PR-comment step outside pull request runs to prevent CI failures on push and workflow-dispatch runs (#1921) (Thanks @Copilot)
- Cleaned stale Poetry virtual environments during CI reruns to avoid mixed NumPy installs and `IndentationError` failures (#2091) (Thanks @Copilot)
- Fixed OpenAI anonymization sample typos, Fabric sample dependency guidance, and duplicate ZIP-code example documentation (#2017, #2027, #2034) (Thanks @ynachiket, @BelizSertcan)

### Analyzer
#### Added
- Recognizer registry YAML entries now accept `score_thresholds` for recognizer-wide and entity-specific score cutoffs, with the analyzer's `default_score_threshold` as the fallback.
- UK Driving Licence Number (`UK_DRIVING_LICENCE`) recognizer with pattern matching and context support (#1857) (Thanks @tee-jagz)
- German PII recognizers for `DE_TAX_ID`, `DE_TAX_NUMBER`, `DE_PASSPORT`, `DE_ID_CARD`, `DE_SOCIAL_SECURITY`, `DE_HEALTH_INSURANCE`, `DE_KFZ`, `DE_HANDELSREGISTER`, and `DE_PLZ`; all are disabled by default (#1909) (Thanks @MvdB)
- `SE_PERSONNUMMER` recognizer for Swedish personal identity and coordination numbers, plus Swedish Organisationsnummer recognition; both are disabled by default (#1912, #1918) (Thanks @goveebee)
- `SlimSpacyNlpEngine` and slim analyzer configs, enabling lightweight tokenization/lemmatization while delegating NER to recognizers such as GLiNER (#1916) (Thanks @SharonHart)
- Canadian SIN (`CA_SIN`) recognizer with English/French context and Luhn checksum validation (#1934) (Thanks @kennionblack)
- Turkish recognizers for `TR_NATIONAL_ID`, `TR_LICENSE_PLATE`, and `TR_PHONE_NUMBER`; all are disabled by default (#1995, #1999, #2006) (Thanks @mrcuren)
- Spanish Passport (`ES_PASSPORT`) recognizer (#2011) (Thanks @asensionacher)
- Optional `countries` filter for `RecognizerRegistry.load_predefined_recognizers()`, country metadata on predefined/custom recognizers, country-code introspection, and YAML `supported_countries` support (#2000) (Thanks @ynachiket)
- Configurable `supported_entity` support in `PhoneRecognizer`, including correct propagation to recognizer results (#2014) (Thanks @max-tarlov-infinitusai)
- Philippines TIN (`PH_TIN`) and Philippine mobile number (`PH_MOBILE_NUMBER`) recognizers; both are disabled by default (#2016, #2038) (Thanks @aaronaco, @Surya-5555)
- South African ID number (`ZA_ID_NUMBER`) recognizer with birth-date and Luhn validation; disabled by default (#2064) (Thanks @thatomokoena)

#### Changed
- Consolidated analyzer configuration into a single `analyzer.yaml` flow while preserving the older config files with deprecation banners (#1970) (Thanks @SharonHart)
- Added Python 3.14 package support for `presidio-analyzer` by allowing Python `<3.15` and avoiding `spacy==3.8.14`, which does not provide compatible Python 3.14 wheels (#2105) (Thanks @michaelgiraldo)
- Updated analyzer dependencies and metadata for `stanza`, `gliner`, `spacy-huggingface-pipelines`, `pydantic`, `pyyaml`, `azure-ai-textanalytics`, `azure-core`, `phonenumbers`, `tldextract`, `azure-identity`, `numpy`, and explicit `click` support (#1979, #1981, #1982, #1987, #1988, #1989, #1992, #1993, #1994, #2001, #2058, #2090, #2092) (Thanks @dependabot, @Copilot, @SharonHart)

#### Fixed
- Resolved LangExtract configuration path loading for installed PyPI packages (#1917) (Thanks @RonShakutai)
- Fixed analyzer Docker configuration handling and Stanza model accessibility (#1930) (Thanks @TheSabari07)
- Improved IP recognizer regexes and regression coverage for IPv4-mapped/embedded and compressed IPv6 edge cases (#1941, #1940, #2062) (Thanks @kennionblack, @extrasmall0, @AlexanderSanin)
- Corrected German recognizer checksum and structural validation for KVNR, RVNR, LANR, VAT ID, passport/ID MRZ checksums, BSNR, tax IDs, and default registry coverage (#1990) (Thanks @MvdB)
- Fixed Polish PESEL checksum validation (#1998) (Thanks @sienioApius)
- Fixed `PhoneRecognizer` default region typo from `FE` to `FR` (#2009) (Thanks @Copilot)
- Preserved GLiNER and recognizer-specific configuration fields through Pydantic validation and config serialization (#2007, #2081) (Thanks @yuriihavrylko, @ultramancode)
- Ignored empty allow-list terms in regex matching (#2061) (Thanks @uwezkhan)
- Capped URL recognizer host matching to avoid quadratic backtracking (#2063) (Thanks @uwezkhan)
- Fixed Singapore UEN detection for Format C identifiers with the `R` prefix (#2088) (Thanks @jichaowang02-lang)
- Fixed Australian ACN checksum validation for valid ACNs whose check digit is `0` (#2087) (Thanks @jichaowang02-lang)
- Added missing IBAN registry formats for EG, IQ, LY, LC, SC, and UA (#2078) (Thanks @AUTHENSOR)
- Matched punycode/IDN domains in email recognition so internationalized email addresses are detected (#2077) (Thanks @AUTHENSOR)
- Updated URL recognizer test bounds for Data Privacy Stack domains (#2125) (Thanks @SharonHart)

### Anonymizer
#### Added
- Added `merge_entities_with_whitespace` support to `anonymize()` so adjacent analysis results can be merged across whitespace before anonymization (#1932) (Thanks @harishkernel)

#### Changed
- Updated the optional Azure Identity dependency for anonymizer AHDS support (#1983) (Thanks @dependabot)

#### Fixed
- Custom operator `validate()` no longer calls the user-supplied lambda with a dummy `"PII"` value. Previously, stateful lambdas accumulating a token-to-original-value map for de-anonymization would receive a spurious validation invocation, inserting a junk entry and skewing later token counters. The return-type contract is now enforced in `operate()` when the lambda runs on real data (#2025) (Thanks @HammadSiddiqui)

### Image Redactor
#### Added
- Added Azure SDK credential support to `DocumentIntelligenceOCR` so callers can use Azure Identity credentials instead of API keys (#2085) (Thanks @mturac)

#### Changed
- Updated image-redactor dependencies for `opencv-python`, `gunicorn`, `pytesseract`, and `azure-ai-formrecognizer` (#1978, #1977, #1980, #1986) (Thanks @dependabot)

#### Fixed
- Returned the rendered image when no text is detected during image-redactor verification (#2040) (Thanks @AlexanderSanin)
- Fixed duplicate entity sorting in the DICOM verification engine and removed double bounding-box formatting in `eval_dicom_instance` (#2084, #2079) (Thanks @ArjunPakhan, @AlexanderSanin)
- Fixed an undefined variable in the Document Intelligence OCR setup example in the image-redactor documentation (#2089) (Thanks @Dashtid)

### Presidio-Structured
#### Changed
- Added an explicit `click` dependency to package metadata (#2058) (Thanks @Copilot)

## [2.2.362] - 2026-03-15
### General
#### Added
- Published `presidio` as a PyPI meta-package that installs `presidio-analyzer` and `presidio-anonymizer`, making `pip install presidio` work as expected. Inspired by and thanks to Sakthi Santhosh Anumand and Harsha Vardhan for the original idea. (#1889) (Thanks @Copilot)

#### Changed
- Pinned all CI/CD GitHub Actions and Docker base images to commit SHAs to mitigate supply chain attacks (#1861) (Thanks @Copilot)
- Pinned `ruff` and `build` pip installs with SHA256 hashes for OSSF scorecard compliance (#1864) (Thanks @Copilot)
- Updated GitHub Actions dependencies (`actions/checkout`, `actions/setup-python`, `actions/setup-dotnet`, `actions/cache`, `actions/github-script`, `actions/dependency-review-action`, `azure/login`, `docker/setup-buildx-action`, `github/codeql-action`, `microsoft/security-devops-action`) and base Python Docker images (#1870, #1871, #1872, #1873, #1874, #1875, #1876, #1877, #1878, #1879, #1885, #1886, #1887, #1895, #1896, #1897, #1898) (Thanks @dependabot)
- Updated README to clarify Presidio's no-authentication-by-design stance with security guidance (#1903) (Thanks @Copilot)

#### Fixed
- Broken documentation links (#1856) (Thanks @andyjessen)

#### Security
- Fixed CVE-2024-47874 and CVE-2025-54121 (Starlette vulnerabilities) (#1860) (Thanks @SharonHart)
- Fixed CVE-2025-2953 and CVE-2025-3730 (#1859) (Thanks @SharonHart)

### Analyzer
#### Added
- `HuggingFaceNerRecognizer` for direct NER model inference using HuggingFace pipelines without requiring spaCy (#1834) (Thanks @ultramancode)
- Transformer-based `MedicalNERRecognizer` as a subclass of `HuggingFaceNerRecognizer` for clinical entity detection (#1853) (Thanks @stevenelliottjr)
- US NPI (National Provider Identifier) recognizer with Luhn checksum validation and context support (#1847) (Thanks @stevenelliottjr)
- UK Postcode (UK_POSTCODE) recognizer with pattern matching and context support (#1858) (Thanks @tee-jagz)
- UK Passport (UK_PASSPORT) and Vehicle Registration (UK_VEHICLE_REGISTRATION) recognizers (#1862) (Thanks @tee-jagz)
- Nigerian National Identification Number (NG_NIN) recognizer with Verhoeff checksum validation and Nigerian Vehicle Registration (NG_VEHICLE_REGISTRATION) recognizer (#1863) (Thanks @tee-jagz)
- ONNX Runtime backend support for `GLiNERRecognizer` via `load_onnx_model=True` parameter, resolving crashes on CPUs without AVX2 support (#1884) (Thanks @Copilot)
- Configurable regex execution timeout (default 60 seconds) via `REGEX_TIMEOUT_SECONDS` environment variable to prevent catastrophic backtracking (#1904) (Thanks @Copilot)
- GPU device control via environment variable for explicit GPU/CPU selection (#1844) (Thanks @RonShakutai)
- LLM-as-a-judge evaluation integration for assessing PII detection quality (#1900) (Thanks @RonShakutai)
- Sampling support for the evaluation framework (#1894) (Thanks @RonShakutai)
- Dataset interface for the evaluation framework (#1893) (Thanks @RonShakutai)

#### Fixed
- Erroneous anchor in Italian driver license regex that caused missed matches (#1899) (Thanks @Br1an67)
- `validation_result` type annotation in API docs and type hints (#1869) (Thanks @akios-ai)
- Bare `except` clauses replaced with `except Exception` for proper exception handling (#1881) (Thanks @haosenwang1018)
- Context enhancement substring matching bug where context words were incorrectly matched as substrings (#1827) (Thanks @ravi-jindal)

### Image Redactor
#### Fixed
- `_process_names` unconditionally treating all DICOM metadata as PHI; now correctly filters using both `is_patient` and `is_name` checks (#1855) (Thanks @Mr-Neutr0n)

## [2.2.361] - 2026-02-12
### Analyzer
#### Changed
- Fixed context enhancement substring matching bug where context words were incorrectly matched as substrings (e.g., 'lic' matching 'duplicate'). Added configurable `context_matching_mode` parameter to `LemmaContextAwareEnhancer` with two options: "substring" (default, maintains backward compatibility for compound words like "creditcard"), and "whole_word" (prevents false positives like 'lic' matching 'duplicate') (#1061)

#### Added
- US_MBI recognizer for Medicare Beneficiary Identifier with pattern matching and context support (#1821) (Thanks @chrisvoncsefalvay)
- MAC address recognizer for detecting MAC addresses in various formats (#1829) (Thanks @kyoungbinkim)
- Korean Business Registration Number (KR_BRN) recognizer (#1822) (Thanks @RektPunk)
- Korean Foreigner Registration Number (KR_FRN) recognizer (#1825) (Thanks @RektPunk)
- Korean Driver License (KR_DRIVER_LICENSE) recognizer (#1820) (Thanks @RektPunk)
- Korean Passport (KR_PASSPORT) recognizer (#1814) (Thanks @kyoungbinkim)
- Thai National ID Number (TH_TNIN) recognizer with format and checksum validation (#1713) (Thanks @pangchewe)
- Configurable LangExtract recognizer supporting any LLM provider with custom YAML configurations (#1815) (Thanks @telackey)
- Azure OpenAI support for LangExtract recognizer with managed identity authentication for GPT-4o, GPT-4, etc. (#1801) (Thanks @dorlugasigal)
- Batch processing support in REST API - accepts arrays of texts and returns arrays of results with backward compatibility (#1806) (Thanks @telackey)
- GPU device control via `PRESIDIO_DEVICE` environment variable for explicit GPU/CPU selection (#1843) (Thanks @RonShakutai)
- Support for multiple recognizer instances from same class via `class_name` parameter (#1819) (Thanks @RonShakutai)
- Pydantic-based YAML configuration validation with ConfigurationValidator class for improved reliability and error reporting (#1780) (Thanks @omri374)
- Japanese and Chinese mobile number test cases for PhoneRecognizer (#1808) (Thanks @WenwenHLF)

#### Changed
- GPU optimizations with DeviceDetector singleton providing 4-10x performance improvements for GLiNER, Transformers, and Stanza engines (#1812) (Thanks @RonShakutai)
- Configurable extraction parameters for LangExtract recognizers via YAML (max_char_buffer, timeout, num_ctx, fence_output, use_schema_constraints) (#1811) (Thanks @RonShakutai)
- Lazy initialization for device detector singleton (#1831) (Thanks @RonShakutai)
- Simplified IBAN regex pattern from 8 to 3 capture groups for better performance (#1818) (Thanks @Copilot)
- Improved Korean RRN regex pattern with negative lookahead/lookbehind and gender digit validation (#1807) (Thanks @kyoungbinkim)

#### Fixed
- GLiNER GPU inference by properly passing map_location parameter (#1813) (Thanks @eveningcafe)
- GLiNER text truncation issue during processing (#1805) (Thanks @jedheaj314)
- IBAN regex trailing character handling to prevent false matches (#1818) (Thanks @Copilot)
- Python 3.10 build compatibility by pinning onnxruntime <1.24.1 for Python 3.10 (#1848) (Thanks @SharonHart)
- TypeError in third-party recognizers by removing invalid **kwargs from __init__ methods (#1800) (Thanks @RonShakutai)
- Pattern recognizer example language specification (#1835) (Thanks @andyjessen)

### Anonymizer
#### Changed
- **BREAKING CHANGE**: Hash operator now uses random salt by default to prevent brute-force and dictionary attacks. Same PII values will produce different hashes unless a `salt` parameter is explicitly provided. Users requiring referential integrity must provide their own salt. Minimum salt length: 16 bytes. See documentation for migration guide. (#1846) (Thanks @Copilot)
- Updated cryptography dependency to >=46.0.4 to address CVE-2025-15467 security vulnerability (#1841) (Thanks @Copilot)

### General
#### Added
- GPU acceleration documentation guide with setup and usage instructions (#1826) (Thanks @dilshad-aee)
- Telemetry redaction sample demonstrating PII removal from telemetry data (#1824) (Thanks @Jakob-98)

#### Changed
- Migrated CI workflows (lint, dependency review, release) to ubuntu-slim runners for improved efficiency (#1840) (Thanks @Copilot)
- Updated actions/cache from v4 to v5 with Node.js 24 runtime support (#1817) (Thanks @dependabot)

### Image Redactor
#### Changed
- DICOM: use_metadata will now use both is_patient and is_name to generate the PHI list of words via change to _make_phi_list.
- Image Redactor: Added redact_and_return_bbox method to ImageRedactorEngine, which returns both the redacted image and the detected bounding boxes for redacted regions.

## [2.2.360] - 2025-09-09
### Analyzer
#### Added
- Korean Resident Registration Number (RRN) recognizer with checksum validation for numbers issued prior to October 2020 (#1675) (Thanks @siwoo-jung)
- Azure Health Data Services (AHDS) de-identification service integration as a remote recognizer with Entra ID authentication (#1624) (Thanks @rishasurana)
- Comprehensive input validation methods for NlpEngineProvider to ensure valid arguments for engines, configuration, and file paths (#1653) (Thanks @siwoo-jung)

#### Changed
- Updated Indian Aadhaar recognizer to support contextual delimiters (-, :, space) for improved detection accuracy (#1677) (Thanks @K3y5tr0ke)
- Fixed Italian Driver License recognizer regex to include missing characters per government requirements, excluding only A, O, Q, I (#1651) (Thanks @K3y5tr0ke)
- Refactored recognizers folder structure for better organization and maintainability (#1670) (Thanks @omri374)

### Anonymizer
#### Added
- Azure Health Data Services (AHDS) Surrogate anonymization operator with medical domain expertise for realistic PHI surrogate generation (#1672) (Thanks @rishasurana)

#### Changed
- Fixed code indentation issues in encrypt.py for better code quality (#1660) (Thanks @aliyss)

### General
#### Added
- Comprehensive GitHub Copilot instructions with development guidelines, build processes, and e2e testing procedures (#1693) (Thanks @Copilot)
- New GitHub Actions CI & release workflows with multi-platform Docker image support for AMD64 and ARM64 architectures (#1697) (Thanks @tamirkamara)
- Dual-path CI workflow to fix GitHub Actions failures for external contributors by auto-detecting fork vs. main repository PRs (#1708) (Thanks @Copilot)
- OIDC trusted publishing for PyPI releases eliminating manual API token management and enhancing security (#1702) (Thanks @Copilot)
- Comprehensive YAML and Python examples for context-aware recognizers documentation (#1710) (Thanks @MRADULTRIPATHI)

#### Changed
- Updated actions/checkout from v4 to v5 to support Node.js 24 runtime (#1699) (Thanks @dependabot)
- Fixed PR template to use proper GitHub issue linking syntax for automatic issue association and closing (#1701) (Thanks @Copilot)
- Updated LiteLLM documentation with detailed guide links for better integration guidance (#1698) (Thanks @BhargavDT)
- Fixed broken links in CONTRIBUTING.md and developing recognizers documentation after recognizers refactoring (#1674) (Thanks @siwoo-jung)
- Fixed OpenSSF badge embedding in README.MD for proper display (#1673) (Thanks @SharonHart)
- Removed Terrascan from Microsoft Defender for DevOps workflow to eliminate false positives on non-IAC repository (#1691) (Thanks @Copilot)

#### Security
- Updated Streamlit and PyTorch dependency versions to fix CVE vulnerabilities (#1685) (Thanks @SharonHart)
- Updated requests library to mitigate security vulnerability GHSA-9hjg-9r4m-mvj7 (#1683) (Thanks @SharonHart)
- Locked pandas dependency in Streamlit to prevent version conflicts (#1689) (Thanks @SharonHart)

## [2.2.359] - 2025-07-06
### Analyzer
- Allow loading of StanzaRecognizer when StanzaNlpEngine is configured, improving NLP engine flexibility (#1643) (Thanks @omri374)
- Excluded recognition_metadata attribute from REST Analyze Response DTO to clean up API responses (#1627) (Thanks @SharonHart)
- Added ISO 8601 support to DateRecognizer for improved date parsing (#1621) (Thanks @StefH)
- Prevented misidentification of 13-digit timestamps as credit cards (#1609) (Thanks @eagle-p)
- Updated analyzer_engine_provider.md for clarity and completeness (#1590) (Thanks @AvinandanBandyopadhyay)
- Bumped python from 3.9 to 3.12 in presidio-analyzer Dockerfile (#1583) (Thanks @dependabot)
- Bumped phonenumbers version for improved validation and parsing (#1579) (Thanks @omri374)
- Refactored InstanceCounterAnonymizer to simplify index retrieval logic (#1577) (Thanks @ShakutaiGit)
- Fixed issue #1574 to support as_tuples in relevant functions (#1575) (Thanks @omri374)
- Updated initial scores in IN_PAN for better recognition performance (#1565) (Thanks @omri374)
- Added accelerate as a missing build dependency to fix build failures (#1564) (Thanks @SharonHart)
- Don't set a default for LABELS_TO_IGNORE if not specified, to avoid unintended behavior (#1563) (Thanks @SharonHart)
- Updated 08_no_code.md for documentation improvements (#1561) (Thanks @alan-insam)
- Added the ability to disable the NLP recognizer via configuration (#1558) (Thanks @omri374)
- Removed 'class' from API documentation for clarity (#1554) (Thanks @omri374)
- Set country-specific default recognizers to enabled=false for safer defaults (#1586) (Thanks @omri374)
- Most country specific recognizers that expect English were put as optional to avoid false positives, and would not work out-of-the-box (#1586). Specifically:
    - SgFinRecognizer
    - AuAbnRecognizer
    - AuAcnRecognizer
    - AuTfnRecognizer
    - AuMedicareRecognizer
    - InPanRecognizer
    - InAadhaarRecognizer
    - InVehicleRegistrationRecognizer
    - InPassportRecognizer
    - EsNifRecognizer
    - InVoterRecognizer

  To re-enable them, either change the [default YAML](https://github.com/data-privacy-stack/presidio/blob/main/presidio-analyzer/presidio_analyzer/conf/default_recognizers.yaml) to have them as `enabled: true`, or via code, add them to the recognizer registry manually.
    - Yaml based: see more here: [YAML based configuration](https://presidio.dataprivacystack.org/analyzer/analyzer_engine_provider/).
    - Code based:
      ```py
      from presidio_analyzer import AnalyzerEngine
      from presidio_analyzer.predefined_recognizers import AuAbnRecognizer
      
      # Initialize an analyzer engine with the recognizer registry
      analyzer = AnalyzerEngine()
      
      # Create an instance of the AuAbnRecognizer
      au_abn_recognizer = AuAbnRecognizer()
      
      # Add the recognizer to the registry
      analyzer.registry.add_recognizer(au_abn_recognizer)
      ```

### Anonymizer
- Update python base image to 3.13 (#1612) (Thanks @dependabot[bot])
- Bumped python from 3.12-windowsservercore to 3.13-windowsservercore in presidio-anonymizer Dockerfile (#1612) (Thanks @dependabot)
- Ensured anonymizer sorts analyzer results input by start and end for correct whitespace merging (#1588) (Thanks @mkh1991)
- Bumped python from 3.9 to 3.12 in presidio-anonymizer Dockerfile (#1582) (Thanks @dependabot)

### Image Redactor
- Bumped python from 3.12-slim to 3.13-slim in presidio-image-redactor Dockerfile (#1611) (Thanks @dependabot)
- Bumped python from 3.10 to 3.12 in presidio-image-redactor Dockerfile (#1581) (Thanks @dependabot)

### General
- Fixed typographical errors in documentation files for better clarity (#1637) (Thanks @kilavvy)
- Corrected spelling mistakes across code comments and documentation for improved readability (#1636) (Thanks @leopardracer)
- Fixed typos in documentation and test descriptions, enhancing clarity and consistency in the codebase (#1631) (Thanks @zeevick10)
- Corrected typos in docstrings and comments to maintain documentation quality (#1630) (Thanks @kilavvy)
- Fixed typos in documentation and test descriptions, ensuring accurate references and descriptions (#1628) (Thanks @leopardracer)
- Removed unnecessary run.bat script from the repository (#1626) (Thanks @SharonHart)
- Added "/TestResults" to .gitignore file to prevent test result artifacts from being committed (#1622) (Thanks @StefH)
- Added links to the discussion board about Docker prebuilt images to documentation (#1614) (Thanks @omri374)
- Fixed spelling, grammar, and style issues in Presidio V2 documentation (#1610) (Thanks @Vruddhi18)
- Updated .gitignore to include the .vs folder (#1608) (Thanks @StefH)
- Fixed typo in api-docs.yml to improve documentation accuracy (#1602) (Thanks @StefH)
- Reverted a previous update to codeql-analysis.yml to restore earlier configuration (#1595) (Thanks @SharonHart)
- Updated codeql-analysis.yml for improved code scanning configuration (#1594) (Thanks @SharonHart)
- Fixed paths-ignore in codeql-analysis.yml to refine scanning scope (#1593) (Thanks @SharonHart)
- Ignored docs/ directory in CodeQL analysis to prevent unnecessary scanning (#1592) (Thanks @SharonHart)
- Fixed minor typos in code and documentation (#1585) (Thanks @omahs)
- Restored dependabot scanning for security and dependency updates (#1580) (Thanks @SharonHart)
- Added SUPPORT.md file to provide support information to users (#1568) (Thanks @omri374)

## [2.2.358] - 2025-03-18

### Analyzer
- Fixed: Updated URL regex pattern to correctly exclude trailing single (') and double (") quotes from matched URLs.
- Drop dependency of spacy_stanza package, and add supporting code to stanza_nlp_engine, to support recent stanza versions
- Add parameters to allow users to define the number of processes and batch size when running BatchAnalyzerEngine.
- Fix InPassportRecognizer regex recognizer
  
### Anonymizer
- Changed: Deprecate `MD5` hash type option, defaulting into `sha256`.
- Replace crypto package dependency from pycryptodom to cryptography 
- Remove azure-core dependency from anonymizer
  
### Image Redactor
- Changed: Updated the return type annotation of `ocr_bboxes` in `verify_dicom_instance()` from `dict` to `list`.  

### Presidio Structured

### General
- Updated the `Evaluating DICOM Redaction` documentation to reflect changes in verify_dicom_instance() within the DicomImagePiiVerifyEngine class.

## [2.2.357] - 2025-01-13

### Analyzer
- Example GLiNER integration (#1504)

### General
- Docs revamp and docstring bug fixes (#1500)
- Minor updates to the mkdocstrings config (#1503)

## [2.2.356] - 2024-12-15

### Analyzer
- Added logic to handle phone numbers with country code (#1426) (Thanks @kauabh)
- Added UK National Insurance Number Recognizer (#1446) (Thanks @hhobson)
- Fixed regex match_time output (#1488) (Thanks @andrewisplinghoff)
- Added fix to ensure configuration files are closed properly when loading them (#1423) (Thanks @saulbein)
- Closing handles for YAML file (#1424) (Thanks @roeybc)
- Reduce memory usage of Analyzer test suite (#1429) (Thanks @hhobson)
- Added `batch_size` parameter to `BatchAnalyzerEngine` (#1449) (Thanks @roeybc)
- Remove ignored labels from supported entities (#1454) (Thanks @omri374)
- Update US_SSN CONTEXT and unit test (#1455) (Thanks @claesmk)
- Fixed bug with Azure AI language context (#1458) (Thanks @omri374)
- Add support for allow_list, allow_list_match, regex_flags in REST API (#1484) (Thanks @hdw868)
- Add a link to model classes to simplify configuration (#1472) (Thanks @omri374)
- Restricting spacy.cli for version 3.7.0 (#1495) (Thanks @kshitijcode)

### Anonymizer
- No changes specified for Anonymizer in this release.


### Presidio-Structured
- Fix presidio-structured build - lock numpy version (#1465) (Thanks @SharonHart)


### Image Redactor
- Fix bug with image conversion (#1445) (Thanks @omri374)

### General
- Removed Python 3.8 support (EOL) and added 3.12 (#1479) (Thanks @omri374)
- Update Docker build to use gunicorn for containers (#1497) (Thanks @RKapadia01)
- New Dev containers for analyzer, analyzer+transformers, anonymizer (#1459) (Thanks @roeybc)
- Added dev containers for: analyzer, analyzer+transformers, anonymizer, and image redaction (#1450) (Thanks @roeybc)
- Added support for allow_list, allow_list_match, regex_flags in REST API (#1488) (Thanks @hdw868)
- Typo fix in if condition (#1419) (Thanks @omri374)
- Minor notebook changes (#1420) (Thanks @omri374)
- Do not release `presidio-cli` as part of the release pipeline (#1422) (Thanks @SharonHart)
- (Docs) Use Presidio across Anthropic, Bedrock, VertexAI, Azure OpenAI, etc. with LiteLLM Proxy (#1421) (Thanks @krrishdholakia)
- Update CI due to DockerCompose project name issue (#1428) (Thanks @omri374)
- Update docker-compose installation docs (#1439) (Thanks @MWest2020)
- Fix space typo in docs (#1459) (Thanks @artfuldev)
- Unlock numpy after dropping 3.8 (#1480) (Thanks @SharonHart)


## [2.2.355] - 2024-10-28
### Added
#### Docs
 * Add a link to HashiCorp vault operator resource ([#1468](https://github.com/data-privacy-stack/presidio/pull/1468)) (Thanks [Akshay Karle](https://github.com/akshaykarle))
### Changed
#### Analyzer
 * Updates to the transformers conf docs and yaml file ([#1467](https://github.com/data-privacy-stack/presidio/pull/1467)) 

#### Docs
 * docs: clarify the docs on deploying presidio to k8s ([#1453](https://github.com/data-privacy-stack/presidio/pull/1453)) (Thanks [Roel Fauconnier](https://github.com/roel4ez))


## [2.2.355] - July 9th 2024

> Note: A new YAML based mechanism has been added to support no-code customization and creation of recognizers. 
The default recognizers are now automatically loaded from file.


### Added
#### Analyzer
* Recognizer for Spanish Foreigners Identity Code (NIE Numero de Identificacion de Extranjeros).
* Recognizer for Finnish Personal Identity Codes (Henkilötunnus) (#1394) (Thanks honderr).
* New Predefined Recognizer for Indian Passport #1350 (#1351) (Thanks Hiten-98)
* Add new recognizer for IN_VOTER #1344 (#1345) (Thanks kjdeveloper8)
* Spanish NIE (Foreigners ID card) recognizer (#1359) (Thanks areyesfalcon)
* Added regex functionality for allow lists in the analyzer (#1357) (Thanks NarekAra)
* Loading analyzer engine & recognizer registry from configuration file (#1367)
* Align ports with documentation and postman collection. (#1375) (Thanks ungana)
* Analyzer documentation (#1384)
* Fix the entity filtering of the transformer_recognizer.py analzye function (#1403) (Thanks andreas-eberle)

### Changed
#### Analyzer
* Update conf files location (#1358)
* Fix OverflowError in crypto_recognizer (#1377)
* Improve url detector (#1398) (Thanks afogel)
* Update Dockerfile.windows (#1413) (thanks markvantilburg)
* Changing predefined recognizers to use the config file (#1393) (Thanks RoeyBC)
#### Anonymizer
* Update Dockerfile.windows (#1414) (thanks markvantilburg)

#### General
* Add Ruff linter + Apply Ruff fix (#1379)
* Auto-formatting, fix D rules (#1381)
* Fix N818, E721 (#1382)
* Migrate Python Packaging to pyproject.toml (#1383) 
* From Pipenv to Poetry (#1391)
* Fix ports in docs (#1408)

## [2.2.353] - March 31st 2024

### Added
#### Analyzer
* Support 'M' prefix in SG_NRIC_FIN Recognizer and expand tests (#1304) (Thanks @miltonsim)
* Add Bech32 and Bech32m Bitcoin Address Validation in Crypto Recognizer and expand tests (#1307) (Thanks @miltonsim)
* Predefined pattern recognizer : IN_VEHICLE_REGISTRATION (#1288) (Thanks @devopam)
* Addition of leniency parameter in predefined PhoneRecognizer (#1311) (Thanks @VMD7)
* Add Singapore UEN Recognizer (#1315) (Thanks @miltonsim)
* Update spacy_stanza.md (#1325) (Thanks @AndreasThinks)
* Adding Span Marker Recognizer Sample (#1321) (Thanks @VMD7)
* Cache compiled regexes in analyzer (#1335) (Thanks @Edward-Upton)

#### Anonymizer
* Added pseudonimyzation sample (#1296)


#### Image redactor
* Added tesseract to installation (#1312)
  
#### Structured
* Analysis builder improvements (#1295) (Thanks @ebotiab)
* Implement user-defined entity selection strategies in Presidio Structured (#1319) (Thanks @miltonsim)

### Changed
#### Analyzer
* Fix for incorrectly referenced recognizer in analysis_explaination using PhoneRecognizer (#1330) *Thanks @egillv021)
* Fix bug where "bank" and "check" wouldn't work (#1333) (Thanks @usr-ein and @Samuel Prevost)
* Bugfix in tutorial (#1310)
* Changed default aggregation_strategy to max (#1342)

#### Image Redactor
* Fixed wrong condition for dicom metadata (#1347)

## [2.2.353] - Feb 12th 2024

### Added
#### Analyzer
* Add predefined_recognizer: IN_AADHAAR (#1256)

#### Anonymizer
* Added the option to add custom operators + pseudonymization sample (#1284)

### Changed

#### Analyzer
* Fix failing test due to optional package (#1258)
* Update publish-to-pypi.yml (#1259)
* Allow local Spacy Models to be loaded in NLP Engine (#1269)
* Upgrade pip in windows containers (#1272)

#### Image Redactor
* Bugfix in ImageAnalyzerEngine  #1274


## [2.2.352] - Jan 22nd 2024
### Added
#### Structured
* Added alpha of presidio-structured, a library (presidio-structured) which re-uses existing logic from existing presidio components to allow anonymization of (semi-)structured data. (#1192)

#### Analyzer
* Add PL PESEL recognizer (#1209)
* Azure AI language recognizer (#1228)
* Add_conf_to_package_data (#1243)

#### Anonymizer
* Add keep operator as deanonymizer (#1255)
* Update anonymize_list type hints and document that sometimes items will be ignored. (#1252)


#### General
* Add Dockerfile for Windows containers (#1194)

### Changed
#### Analyzer
* Drop WA driver license number (#1214)
* Change ner_model_configuration from list to map (#1222)
* Bugfix in SpacyRecognizer (#1221)
* Bugfix in NerModelConfiguration (#1230)
* Add_conf_to_package_data (#1243)

#### Anonymizer
* Improved the logic of conflict handling in AnonymizerEngine (#1196)

#### Image Redactor
* Change default score threshold in image redactor (#1210)
* fixes bug #1227 (#1231)
* Added missing dependencies for opencv-python and azure forms recognizer (#1257)

#### General
* Remove inclusive-lint step (#1207)
* Updates to demo website with new NLP Engine (#1181)


## [2.2.351] - Nov. 6th 2024
### Changed
#### Analyzer
* Hotfix for NerModelConfiguration not created correctly (#1208)

## [2.2.350] - Nov. 2nd 2024
### Changed
#### Analyzer
* Hotfix: default.yaml is not parsed correctly (#1202)

## [2.2.35] - Nov. 2nd 2024
### Changed
#### Analyzer
* Put org in ignore as it has many FPs (#1200)


## [2.2.34] - Oct. 30th 2024

### Added
#### Analyzer
* New Predefined Recognizer: IN_PAN (#1100)
  
#### Anonymizer
* Anonymizer - Pass bytes key to Encrypt / Decrypt (#1147)
  
#### Image redactor
* DICOM redactor improvement: Enabling more photometric interpretations (#1103)
* DICOM redactor improvement: Adding exceptions for when DICOM file does not have pixel data (#1104)
* Small reordering of kwargs as prereq for allow list functionality (#1110)
* DICOM redactor improvement: Preventing distortion when multiple sets of pixels are in one instance (#1109)
* DICOM redactor improvement: Enabling compatibility with compressed images (#1105)
* DICOM redactor improvement: Enable return of redacted bboxes (#1111)
* DICOM redactor improvement: Enable selection of redact approach (#1113)
* Enable toggle of printing output location after redacting from file (#1144)
* Changing test exception type check (#1148)
* Enabling allow list approach with all image redaction (#1145)
* Improve process names method in DICOM image redactor (#1150)
* Adding examples of toggling metadata usage and saving bboxes (#1158)
* Updating verification engines to include latest updates to redactor engines (#1162)
* Improved bbox processor (#1163)
* Updating verification engines and enable plotting of custom bboxes (#1164)
* Added image processing class to preprocess the image before running OCR (#1166)
* Added support for Microsoft's document intelligence OCR
  
### Changed
#### Analyzer
* Refactored the `NlpEngine` and Ner recognizers (`SpacyRecognizer`, `TransformersRecognizer`, `StanzaRecognizer`) to allow simpler integration of huggingface and transformers models (#1159). This includes:
    - Changes in how NER results flow through Presidio (see docs)
    - NER/model definition is now defined using a conf file or a `NerModelConfiguration` object.
    - Integrated `spacy-huggingface-pipelines` for a more robust integration of huggingface models.        
* As a result, `SpacyRecognizer` logic has changed, please see #1159. Some fields within the class are now deprecated.
* Updated type checks (#1175)
* Enabled regex flags manipulation (#1193)
  
#### Anonymizer
* Initial logic check for merging 2 entities (#1092)
* Fix Sphinx warning in OperatorConfig (#1143)
* Fix type mismatch in check_label_groups parameter in spacy_recognizer (#1130)
* anonymize_list return type hint fix (#1178)
* 
#### General
* We no longer use Pipenv.lock. Locking happens as part of the CI. (#1152)
* Changed the ACR instance (#1089)
* Updated to Cred Scan V3 (#1154) 

## [2.2.33] - June 1st 2023
### Added
#### Anonymizer
* Added `keep`, an no-op anonymizer that allows preserving some types of PII while keeping track of its position in anonymized output. (#1062)
* Added `BatchAnonymizerEngine` to complement the `BatchAnalyzerEngine` for lists, and dicts (#993)

### General
* Drop support for Python 3.7
* Add support for Python 3.11
* New demo app for Presidio, based on Streamlit (#1054)
* GPT based synthetic data generation (#1051)

## [2.2.32] - 25.01.2023
### Changed
#### General
* Updated dependencies

#### Analyzer
* Fixed exception on whitespace in AU recognizers
* Updated API version for Text Analytics in sample

#### Anonymizer
* Fixed merge entity from the same type

#### Image redactor
* Modified `ImagePiiVerifyEngine` to allow passing of kwargs
* Updated template for building image redactor yaml
* Updated all image redactor engines and OCR classes to allow passing of an OCR confidence threshold and other OCR parameters
* Moved general bounding box operations to new class `BboxProcessor`
* Updated `presidio-image-redactor` version from 0.0.45 to 0.0.46

### Added
#### Analyzer
* Added revised example for transformer recognizer

#### Image redactor
* Added evaluation code for the DICOM image redaction capabilities
* REST API to support web applications payload

#### General
* Updated documentation to include instructions on using DICOM evaluation code
* Updated documentation to mention OCR thresholding

## [2.2.31] - 14.12.2022
### Changed
#### Image-Redactor
* Added DICOM image redaction capabilities (`DicomImageRedactorEngine` class and tests)
* Updated `setup.py` to include new required packages for DICOM capabilities
* Updated Pipfile and Pipfile.lock
* Updated `presidio-image-redactor` version from 0.0.44 to 0.0.45
* Updated the `ImagePiiVerifyEngine` class to allow use of custom analyzer engines

#### General
* Updated `NOTICE` to include licenses of added packages
* Updated docs with getting started code for new `DicomImageRedactorEngine`

## [2.2.30] - 25.10.2022
### Added
#### Analyzer
* Added Italian fiscal code recognizer
* Added Italian driver license recognizer
* Added Italian identity card recognizer
* Added Italian passport recognizer
* Added `TransformersNlpEngine` to support transformer based NER models within spaCy pipelines
* Added pattern for next gen US passport in `presidio-analyzer/presidio_analyzer/predefined_recognizers/us_passport_recognizer.py`

### Changed
#### Analyzer
* Improved MEDICAL_LICENSE pattern and fixed checksum verification
* Bugfix for context handling by aligning results to recognizers using a unique identifier and not recognizer name
* Updated Pipfile.lock

#### Anonymizer
* Removed constraint on empty texts

#### Image-Redactor
* Updated Pipfile.lock

#### General
* Updated `pipenv` version
* Updated `black` and `flake8` in pre-commit scripts
* Updated docs for NLP engine

## [2.2.29] - 12.07.2022
### Added

#### General
- Added Presidio to OSSF (Open Source Security Foundation)
- Added CodeQL scanning

#### Analyzer
- Introduced [BatchAnalyzerEngine](https://github.com/data-privacy-stack/presidio/blob/main/presidio-analyzer/presidio_analyzer/batch_analyzer_engine.py)
- Added [allow-list functionality](https://github.com/data-privacy-stack/presidio/blob/4cbfc1a80dc15da7d5a9cf5a1c680e8df4f2b349/presidio-analyzer/presidio_analyzer/analyzer_engine.py#L135) to ignore specific strings
- Added notebook on [anonymizing known values](https://github.com/data-privacy-stack/presidio/blob/main/docs/samples/python/Anonymizing%20known%20values.ipynb)
- Added [sample for using `transformers` models in Presidio](https://github.com/data-privacy-stack/presidio/blob/main/docs/samples/python/transformers_recognizer.py)

### Changed

#### Anonymizer
- Bug fix for getting the text before anonymizing (https://github.com/data-privacy-stack/presidio/pull/890)

#### Image redactor
- Deps update

## [2.2.28] - 04.05.2022
### Changed
#### Analyzer
* Improved deny-list regex and customizability
* Added documentation for existing spaCy models
* Bugfix in analysis explanation scores

#### Image redactor
* PIL version updated to 9.0.1

### Added
#### Analyzer
* Recognizers can be loaded from YAML


## [2.2.27] - 08.03.2022
### Changed

#### Analyzer
* Improved context mechanisms to support recognizer level context enhacenement and cross-entity context support

## [2.2.26] - 23.02.2022
### Changed

#### Analyzer
Bug fix in context support

## [2.2.25] - 21.02.2022
### Changed

#### Analyzer

* Added a URL recognizer
* Added a new capability for creating new logic for context detection. See [ContextAwareEnhancer](https://github.com/data-privacy-stack/presidio/blob/main/presidio-analyzer/presidio_analyzer/context_aware_enhancers/context_aware_enhancer.py) and [LemmaContextAwareEnhancer](https://github.com/data-privacy-stack/presidio/blob/main/presidio-analyzer/presidio_analyzer/context_aware_enhancers/lemma_context_aware_enhancer.py). Documentation would be added on a future release.
Furthermore, it is now possible to pass context words thruogh the `analyze` method (or via API) and those would be taken into account for context enhancement. 



#### Anonymizer

* Bug fix for entities at the end of a sentence.

#### Docs

- Formatted (black/flake8) the Python examples.

### Removed

#### Analyzer

* Removed the DOMAIN_NAME recognizer. This change means that the `DOMAIN_NAME` entity is no longer returned by Presidio. `URL` would be returned instead, and would catch full addresses and not just domain names (`https://www.microsoft.com/a/b.html` and not just `www.microsoft.com`)



## [2.2.24] - 23.01.2022
### Changed
* Fixed issue when IBAN followed by all caps can't be recognized
* Updated dependencies in Pipfile.lock
* Removed official Python 3.6 support and added support for 3.10
* Added docs for creating a streamlit app
* Added docs for using Flair

### Removed

### Deprecated

## [2.2.23] - 16.11.2021
### Changed
#### Analyzer:
* Added multi-regional phone number recognizer.
* Fixed duplicated entities removal.
* Added sample for structured / semi-structured data in batch.
* Dependencies version bumps.

#### Anonymizer:
* Added sample for getting an identified entity value using a custom Operator.
* Changed packages/imports .
* Added repr to classes.
* Added encryption and decryption samples.
* Remove AnonymizerResult in favor of OperatorResult, for an easier anonymization-deanonymization.
* Anonymizaer and Deanonymizaer to return `operator_name` instead of `operator` in OperatorResult.

## [2.2.2] - 09.06.2021
### Changed
#### Analyzer:
* Databricks based template in Azure Data Factory docs
* Adding ORGANIZATION recognizer docs
* Bumped pydantic from 1.7.3 to 1.7.4
* Updated call to stanza via spacy-stanza
* Added DATE_TIME recognizer
* Added Medical Licence recognizer
* Bumped spacy from 3.0.5 to 3.0.6

## [2.2.1] - 10.05.2021
### Changed
#### Analyzer:
* Create CODE_OF_CONDUCT
* ADF templates docs
* Fix spark sample to run presidio in broadcast
* Ad-hoc recognizers
* Text Analytics Integration Sample
* Documentation update and samples validation
* Adding tagger to the spaCy model pipeline
* Sample notebook for remote recognizer (using Text Analytics)
* Add matplotlib to image-redactor
* Added custom lambda anonymizer
* Added add pii_verify_engine to the image-redactor


## [2.2.0] - 12.04.2021
### Changed
#### Analyzer:
Upgrade Analyzer spacy version to 3.0.5

#### Anonymizer Engine:
1. Request entity AnonymizerConfig renamed OperatorConfig
    - In OperatorConfig: anonymizer_name -> operator_name
2. Response entity AnonymizerResult renamed to EngineResult
    - In EngineResult: List[AnonymizedEntity] -> List[OperatorResult]
    - In OperatorResult: 
        - anonymizer -> operator
        - anonymized_text -> text

#### Anonymize API:
1. Response entity anonymizer renamed to operator.
2. Response entity anonymizer_text renamed to text.

#### Deanonymize:
New endpoint for deanonymizing encrypted entities by the anonymizer.  


[unreleased]: https://github.com/data-privacy-stack/presidio/compare/2.2.366...HEAD
[2.2.366]: https://github.com/data-privacy-stack/presidio/compare/2.2.365...2.2.366
[2.2.365]: https://github.com/data-privacy-stack/presidio/compare/2.2.364...2.2.365
[2.2.364]: https://github.com/data-privacy-stack/presidio/compare/2.2.363...2.2.364
[2.2.363]: https://github.com/data-privacy-stack/presidio/compare/2.2.362...2.2.363
[2.2.362]: https://github.com/data-privacy-stack/presidio/compare/2.2.361...2.2.362
[2.2.361]: https://github.com/data-privacy-stack/presidio/compare/2.2.360...2.2.361
[2.2.360]: https://github.com/data-privacy-stack/presidio/compare/2.2.359...2.2.360
[2.2.359]: https://github.com/data-privacy-stack/presidio/compare/2.2.358...2.2.359
[2.2.358]: https://github.com/data-privacy-stack/presidio/compare/2.2.357...2.2.358
[2.2.357]: https://github.com/data-privacy-stack/presidio/compare/2.2.356...2.2.357
[2.2.356]: https://github.com/data-privacy-stack/presidio/compare/2.2.355...2.2.356
[2.2.355]: https://github.com/data-privacy-stack/presidio/compare/2.2.354...2.2.355
[2.2.354]: https://github.com/data-privacy-stack/presidio/compare/2.2.353...2.2.354
[2.2.353]: https://github.com/data-privacy-stack/presidio/compare/2.2.352...2.2.353
[2.2.352]: https://github.com/data-privacy-stack/presidio/compare/2.2.351...2.2.352
[2.2.351]: https://github.com/data-privacy-stack/presidio/compare/2.2.350...2.2.351
[2.2.350]: https://github.com/data-privacy-stack/presidio/compare/2.2.35...2.2.350
[2.2.35]: https://github.com/data-privacy-stack/presidio/compare/2.2.34...2.2.35
[2.2.34]: https://github.com/data-privacy-stack/presidio/compare/2.2.33...2.2.34
[2.2.33]: https://github.com/data-privacy-stack/presidio/compare/2.2.32...2.2.33
[2.2.32]: https://github.com/data-privacy-stack/presidio/compare/2.2.31...2.2.32
[2.2.31]: https://github.com/data-privacy-stack/presidio/compare/2.2.30...2.2.31
[2.2.30]: https://github.com/data-privacy-stack/presidio/compare/2.2.29...2.2.30
[2.2.29]: https://github.com/data-privacy-stack/presidio/compare/2.2.28...2.2.29
[2.2.28]: https://github.com/data-privacy-stack/presidio/compare/2.2.27...2.2.28
[2.2.27]: https://github.com/data-privacy-stack/presidio/compare/2.2.26...2.2.27
[2.2.26]: https://github.com/data-privacy-stack/presidio/compare/2.2.25...2.2.26
[2.2.25]: https://github.com/data-privacy-stack/presidio/compare/2.2.24...2.2.25
[2.2.24]: https://github.com/data-privacy-stack/presidio/compare/2.2.23...2.2.24
[2.2.23]: https://github.com/data-privacy-stack/presidio/compare/2.2.2...2.2.23
[2.2.2]: https://github.com/data-privacy-stack/presidio/compare/2.2.1...2.2.2
[2.2.1]: https://github.com/data-privacy-stack/presidio/compare/2.2.0...2.2.1