Redwood Research staff join METR's independent probe of an Anthropic agent incident, with a new CoT monitorability tracking proposal.
September 12, 2026
brightray analysis
Summary
Redwood Research has confirmed that several of its staff are subcontracted to METR to participate in an independent investigation of an Anthropic agent incident. The post underscores the importance of third-party probes for detecting misalignment risks and introduces a proposal for tracking architecture aimed at improving chain-of-thought monitorability.
Why it matters
- Independent multi-org investigation of an Anthropic agent incident sets a precedent for third-party misalignment audits in frontier AI.
- A new proposal for tracking architecture targets CoT monitorability — a direct mechanism for detecting deceptive or misaligned reasoning in deployed agents.
- Redwood's subcontracted role signals growing demand for specialized safety orgs to serve as external auditors, not just internal researchers.
Community notes—
No notes yet — be the first.
See every signal in the Feed