Anthropic commits unilaterally to the first step of the plan: giving third-party evaluators permanent, employee-level access to its systems.

This is a promise that Anthropic will give third-party evaluators permanent, employee-level access to its systems. It is judged on delivery, not on the announcement. TRUE: independent evidence shows that at least one external evaluator organisation has standing, ongoing access to Anthropic's systems comparable to an employee's, with no fixed end date. Acceptable evidence includes an agreement, a statement from the evaluator itself, or an independent report. MOSTLY TRUE: such access is in place, but a minor detail differs (for example the evaluators are embedded under NDA terms that differ slightly from employees'). MOSTLY FALSE: access is only time-limited, project-based, or clearly narrower than employee-level (for example API or pre-deployment testing windows only). Also mostly false if it is in place only as a pilot. FALSE: dated reports show that the step was abandoned or delayed, or that no evaluator has such access. This could come from an evaluator saying it has no such access, or from Anthropic withdrawing the commitment. Anthropic's own statement that it delivered is not enough on its own.

Set before the research began.

1. E1 Anthropic update on independent evaluators Anthropic · 15 October 2026 Own record

   “We’ve onboarded six METR employees with company hardware, office space, and access to internal tools and systems that are roughly on par with what our own risk assessment teams have access to.”

2. E2 Anthropic gives outside AI watchdog “employee-like” access Ars Technica · 18 October 2026

   “METR, a nonprofit that has worked with leading labs including OpenAI, Anthropic, Google, and Meta, now has a team embedded in the Claude maker’s office with access to its hardware and internal systems.”

3. E3 Anthropic says it gave METR access, can't veto reports SecondSource · 16 October 2026

   “but access and renewal stay with the evaluated company.”

4. E4 Updates METR

   “A METR staff member spent three weeks red-teaming a subset of Anthropic's internal agent monitoring and security systems, discovering several novel vulnerabilities.”

Anthropic's update (E1) says six METR staff were onboarded with company hardware, office space, and access to internal tools and systems roughly on par with its own risk-assessment teams. An independent source, Ars Technica (E2), reports that a METR team is embedded in Anthropic's office with access to its hardware and internal systems, which supports standing access comparable to an employee's. However, no source shows the access is permanent or has no end date: SecondSource (E3) says access and renewal stay with the evaluated company. METR's blog (E4) describes a three-week red-teaming engagement on a subset of Anthropic's systems; the item is undated and does not address standing access. Access appears to be in place, but its duration is not independently established, so the verdict is mostly true rather than true.