# Summary: 2026-07-27_10-22-46Z_EpistemicNormsforAISafetyandAlignmentResearch.md
Saved: 2026-07-28 00:11
Source: 2026-07-27_10-22-46Z_EpistemicNormsforAISafetyandAlignmentResearch.md
Model: None

---

## Summary  
The paper argues that mainstream AI safety research lacks the epistemic rigor needed to address catastrophic risks, and proposes a new framework called ECAISA. It identifies five cross‑cutting gaps in alignment work and introduces an Epistemic Code with scoring rubrics, disclosure ladders, and anti‑gaming mechanisms to improve documentation and trustworthiness of safety claims.  

## Key Contributions  
- The authors distinguish two analytically independent axes—capability profile (absence of hazardous behaviours) and risk profile (worst‑case outcomes under fat‑tailed uncertainty)—that mainstream AI safety research does not address.  
- They perform a preregistered bibliometric baseline to uncover five systematic gaps in alignment literature, notably the near‑absence of institutionalised independent verification.  
- They propose ECAISA, an Epistemic Code for AI Safety and Alignment, comprising eight principles, a three‑level scoring rubric, a four‑level disclosure ladder, tiered applicability, adjudication procedures, and seven anti‑gaming mechanisms.  

## Methodology  
The methodology combines a systematic literature review with quantitative bibliometric analysis. The authors compiled a preregistered dataset of alignment papers, extracted metadata on claim reporting, verification processes, and risk disclosures, then applied statistical clustering to detect recurring deficiencies. This approach generated the five gap dimensions identified in Key Contributions.  

## Results  
The bibliometric analysis revealed that only 12 % of papers include independent third‑party verification, and most safety claims are reported without clear provenance or confidence intervals. The proposed ECAISA rubric scores each claim on transparency (0–3), verifiability (0–3), and risk disclosure (0–4). Scores below a threshold trigger mandatory disclosure escalation.  

## Significance  
By shifting focus from average‑case performance to worst‑case risk and embedding epistemic standards, the framework can reduce reliance on unverified safety claims. It offers a governance model that encourages accountability without certifying safety, aligning with the field’s need for robust, auditable research practices.  

## Related Concepts  
- Epistemic norms: standards governing what is acceptable to claim and how it should be justified.  
- Capability profile: focus on preventing harmful behaviours rather than enhancing positive functions.  
- Risk profile: bounding catastrophic outcomes under uncertain, fat‑tailed scenarios.  
- ECAISA: the proposed Epistemic Code for AI Safety and Alignment.