generated: '2026-09-12'
method: searched
source: >-
  https://github.com/Project-HAMi/HAMi/blob/master/SECURITY.md and
  https://github.com/Project-HAMi/HAMi/blob/master/SECURITY-INSIGHTS.yml (saved verbatim to
  security/hami-security-insights.yml)
note: >-
  0-working/probe-security-programs.py found nothing because HAMi serves no /.well-known/security.txt
  (404 on project-hami.io, www.project-hami.io and project-hami.github.io) and runs no bug bounty. The
  program is real, it is just published where an open-source project publishes it: a SECURITY.md in the
  repository root, machine-readably cross-declared in an OpenSSF Security Insights 2.0.0 document. Both
  were fetched for this record.
program_published: true
security_txt: false
security_txt_note: >-
  Not served on any HAMi host. Adding one at https://project-hami.io/.well-known/security.txt pointing
  at SECURITY.md would make this program machine-discoverable; it is the single clearest gap in an
  otherwise well-published security posture.
policy_url: https://github.com/Project-HAMi/HAMi/blob/master/SECURITY.md
reporting:
  channel: GitHub Security Advisories (private vulnerability reporting)
  url: https://github.com/Project-HAMi/HAMi/security/advisories/new
  reports_accepted: true
  public_disclosure_before_process: discouraged
  contact_name: HAMi maintainers
  information_requested:
    - Clear and concise description of the vulnerability
    - Steps to reproduce
    - Potential attack scenarios or security impact
    - Suggested mitigations or fixes, if available
bug_bounty:
  available: false
  evidence: >-
    SECURITY.md states "Currently, HAMi does not offer a public bug bounty program"; SECURITY-INSIGHTS.yml
    sets vulnerability-reporting.bug-bounty-available to false.
response_target:
  detail: >-
    Maintainers aim to reply as soon as possible, ideally within 5 working days, allowing for weekends,
    holidays and time-zone differences.
scope:
  in_scope:
    - A report that lets a workload reach another tenant's data, device, or namespace it was not granted.
  out_of_scope:
    - >-
      A report that a workload can exceed its own quota without affecting another tenant. HAMi states
      plainly that in-container enforcement (HAMi-core/libvgpu) limits GPU memory and compute for
      cooperative multi-tenant sharing on a TRUSTED cluster and "is not a hard security boundary
      against a workload with enough privilege to bypass its own hook, for example by unsetting
      LD_PRELOAD, using a static binary, or ptrace". That threat-model statement is unusually explicit
      and is the most useful single sentence in the policy for anyone evaluating HAMi as isolation.
supported_versions:
  cross_ref: lifecycle/hami-lifecycle.yml
security_tooling:
  source: security/hami-security-insights.yml
  tools:
    - {name: Dependabot, type: SCA, ci: true}
    - {name: CodeQL, type: SAST, ci: true}
    - {name: GitHub secret scanning, type: secret-scanning, ci: true, note: push protection enabled}
  additional: FOSSA license and dependency checks, and container image scanning, per the self-assessment comment.
assessments:
  - name: CNCF general technical review
    url: https://github.com/Project-HAMi/HAMi/blob/master/docs/general-technical-review.md
  - name: Audit report (v2.9.0)
    url: https://github.com/Project-HAMi/HAMi/blob/master/docs/audit-report-v290.md
trust_center: false
certifications: []