OpenAI Signs Agreements to Embed AI Evaluators
Source: The Information (opens in a new tab) · Tiffany Li
Intel Summary
OpenAI has signed agreements granting external research organizations—Apollo Research, METR, Transluce, and Goodfire—access to internal information to assess risks across its technology and practices, operationalizing previous commitments to embed independent evaluators.
Why It Matters
Granting outside organizations direct access for risk evaluation establishes a structured third-party auditing mechanism for frontier AI systems, setting a potential standard for transparency and safety governance among proprietary model developers.
Organizations & Entities
Related Intelligence
- ReportAlso involving METR
Anthropic grants non-profit METR ‘employee-like’ access to monitor AI risks
Anthropic has granted non-profit evaluation group METR internal, employee-level access to monitor and evaluate artificial intelligence risks. The arrangement aligns with CEO Dario Amodei's proposal to incorporate independent external evaluators to assess model safety concerns.
Financial Times (AI) - ReportAlso involving METR
Anthropic is letting METR researchers look under the hood with employee-level access
TechCrunch reports that Anthropic is granting researchers from METR employee-level internal access to assess its frontier AI risks, incidents, and internal practices. Under the agreement, METR will independently publish its assessments without Anthropic retaining the right to block their release.
TechCrunch - DevelopmentNewAlso involving METR
Anthropic grants METR researchers employee-level access for AI safety assessments
TechCrunch reports that Anthropic is granting researchers from METR employee-level internal access to assess its frontier AI risks, incidents, and internal practices. Under the agreement, METR will independently publish its assessments without Anthropic retaining the right to block their release. Claims are as reported; this summary makes no determination about accuracy or significance.
- DevelopmentNewAlso involving METR
Audit by METR and Redwood Research reveals safety monitoring gaps at OpenAI
External auditors from METR and Redwood Research completed a three-week on-site evaluation of OpenAI's internal monitoring systems for AI agents, discovering that up to 25 percent of agent traffic from models at the GPT-5.5 class and above went unmonitored. OpenAI published only an executive summary of the 61-page report, with large portions redacted. Claims are as reported; this summary makes no determination about accuracy or significance.