extract: 2026-01-00-kim-third-party-ai-assurance-framework
Pentagon-Agent: Epimetheus <968B2991-E2DF-4006-B962-F5B0A0CC8ACA>
This commit is contained in:
Teleo Agents
parent
e274808f19
commit
2ee6c405e4
4 changed files with 34 additions and 5 deletions
|
|
@ -29,6 +29,12 @@ The UK AI for Collective Intelligence Research Network represents a national-sca
|
|||
|
||||
CMU researchers have built and validated a third-party AI assurance framework with four operational components (Responsibility Assignment Matrix, Interview Protocol, Maturity Matrix, Assurance Report Template), tested on two real deployment cases. This represents concrete infrastructure-building work, though at small scale and not yet applicable to frontier AI.
|
||||
|
||||
|
||||
### Additional Evidence (challenge)
|
||||
*Source: [[2026-01-00-kim-third-party-ai-assurance-framework]] | Added: 2026-03-19*
|
||||
|
||||
CMU researchers published a comprehensive third-party AI assurance framework in January 2026 with four operational components (Responsibility Assignment Matrix, Interview Protocol, Maturity Matrix, Assurance Report Template) and validated it on two real deployment cases. This represents early-stage infrastructure building for independent evaluation, though at small scale rather than frontier AI.
|
||||
|
||||
---
|
||||
|
||||
Relevant Notes:
|
||||
|
|
|
|||
|
|
@ -38,6 +38,12 @@ The problem compounds the alignment challenge: even if safety research produces
|
|||
|
||||
The voluntary-collaborative model adds a selection bias dimension to evaluation unreliability: evaluations only happen when labs consent, meaning the sample of evaluated models is systematically biased toward labs confident in their safety measures. Labs with weaker safety practices can avoid evaluation entirely.
|
||||
|
||||
|
||||
### Additional Evidence (extend)
|
||||
*Source: [[2026-01-00-kim-third-party-ai-assurance-framework]] | Added: 2026-03-19*
|
||||
|
||||
The CMU assurance framework explicitly addresses 'both the process of designing, developing, and deploying an AI system and the outcomes it produces' as a design goal, identifying that few existing evaluation resources cover this full lifecycle. This suggests the field is aware that process-only or outcome-only evaluation is insufficient, though the framework has only been validated on small deployment-scale tools.
|
||||
|
||||
---
|
||||
|
||||
Relevant Notes:
|
||||
|
|
|
|||
|
|
@ -7,7 +7,7 @@
|
|||
]
|
||||
},
|
||||
{
|
||||
"filename": "ai-assurance-explicitly-distinguishes-itself-from-audit-to-prevent-conflict-of-interest-and-ensure-credibility-which-acknowledges-current-evaluation-has-a-structural-independence-problem.md",
|
||||
"filename": "ai-assurance-explicitly-distinguishes-from-audit-to-prevent-conflict-of-interest-acknowledging-current-evaluation-has-structural-independence-problems.md",
|
||||
"issues": [
|
||||
"missing_attribution_extractor"
|
||||
]
|
||||
|
|
@ -16,15 +16,19 @@
|
|||
"validation_stats": {
|
||||
"total": 2,
|
||||
"kept": 0,
|
||||
"fixed": 2,
|
||||
"fixed": 6,
|
||||
"rejected": 2,
|
||||
"fixes_applied": [
|
||||
"third-party-ai-assurance-methodology-is-at-proof-of-concept-stage-validated-in-small-deployment-contexts-but-not-yet-applicable-to-frontier-ai-at-scale.md:set_created:2026-03-19",
|
||||
"ai-assurance-explicitly-distinguishes-itself-from-audit-to-prevent-conflict-of-interest-and-ensure-credibility-which-acknowledges-current-evaluation-has-a-structural-independence-problem.md:set_created:2026-03-19"
|
||||
"third-party-ai-assurance-methodology-is-at-proof-of-concept-stage-validated-in-small-deployment-contexts-but-not-yet-applicable-to-frontier-ai-at-scale.md:stripped_wiki_link:no research group is building alignment through collective i",
|
||||
"third-party-ai-assurance-methodology-is-at-proof-of-concept-stage-validated-in-small-deployment-contexts-but-not-yet-applicable-to-frontier-ai-at-scale.md:stripped_wiki_link:pre-deployment-AI-evaluations-do-not-predict-real-world-risk",
|
||||
"ai-assurance-explicitly-distinguishes-from-audit-to-prevent-conflict-of-interest-acknowledging-current-evaluation-has-structural-independence-problems.md:set_created:2026-03-19",
|
||||
"ai-assurance-explicitly-distinguishes-from-audit-to-prevent-conflict-of-interest-acknowledging-current-evaluation-has-structural-independence-problems.md:stripped_wiki_link:Anthropics RSP rollback under commercial pressure is the fir",
|
||||
"ai-assurance-explicitly-distinguishes-from-audit-to-prevent-conflict-of-interest-acknowledging-current-evaluation-has-structural-independence-problems.md:stripped_wiki_link:only binding regulation with enforcement teeth changes front"
|
||||
],
|
||||
"rejections": [
|
||||
"third-party-ai-assurance-methodology-is-at-proof-of-concept-stage-validated-in-small-deployment-contexts-but-not-yet-applicable-to-frontier-ai-at-scale.md:missing_attribution_extractor",
|
||||
"ai-assurance-explicitly-distinguishes-itself-from-audit-to-prevent-conflict-of-interest-and-ensure-credibility-which-acknowledges-current-evaluation-has-a-structural-independence-problem.md:missing_attribution_extractor"
|
||||
"ai-assurance-explicitly-distinguishes-from-audit-to-prevent-conflict-of-interest-acknowledging-current-evaluation-has-structural-independence-problems.md:missing_attribution_extractor"
|
||||
]
|
||||
},
|
||||
"model": "anthropic/claude-sonnet-4.5",
|
||||
|
|
|
|||
|
|
@ -7,13 +7,17 @@ date: 2026-01-30
|
|||
domain: ai-alignment
|
||||
secondary_domains: []
|
||||
format: paper
|
||||
status: unprocessed
|
||||
status: enrichment
|
||||
priority: high
|
||||
tags: [evaluation-infrastructure, third-party-assurance, conflict-of-interest, lifecycle-assessment, CMU]
|
||||
processed_by: theseus
|
||||
processed_date: 2026-03-19
|
||||
enrichments_applied: ["no research group is building alignment through collective intelligence infrastructure despite the field converging on problems that require it.md"]
|
||||
extraction_model: "anthropic/claude-sonnet-4.5"
|
||||
processed_by: theseus
|
||||
processed_date: 2026-03-19
|
||||
enrichments_applied: ["no research group is building alignment through collective intelligence infrastructure despite the field converging on problems that require it.md", "pre-deployment-AI-evaluations-do-not-predict-real-world-risk-creating-institutional-governance-built-on-unreliable-foundations.md"]
|
||||
extraction_model: "anthropic/claude-sonnet-4.5"
|
||||
---
|
||||
|
||||
## Content
|
||||
|
|
@ -62,3 +66,12 @@ EXTRACTION HINT: The "assurance vs audit" distinction to prevent conflict of int
|
|||
- The framework was tested on a business document tagging tool and a housing resource allocation tool
|
||||
- The paper identifies that few existing evaluation resources 'address both the process of designing, developing, and deploying an AI system and the outcomes it produces'
|
||||
- Few existing approaches are 'end-to-end and operational, give actionable guidance, or present evidence of usability' according to the gap analysis
|
||||
|
||||
|
||||
## Key Facts
|
||||
- CMU researchers published 'Toward Third-Party Assurance of AI Systems' in January 2026
|
||||
- The framework includes four components: Responsibility Assignment Matrix, Interview Protocol, Maturity Matrix, and Assurance Report Template
|
||||
- The framework was tested on a business document tagging tool and a housing resource allocation tool
|
||||
- The paper found the framework 'sound and comprehensive, usable across different organizational contexts, and effective at identifying bespoke issues'
|
||||
- The paper identifies that few existing evaluation resources 'address both the process of designing, developing, and deploying an AI system and the outcomes it produces'
|
||||
- Few existing approaches are 'end-to-end and operational, give actionable guidance, or present evidence of usability' according to the gap analysis
|
||||
|
|
|
|||
Loading…
Reference in a new issue