Anthropic gives outside AI watchdog 'employee-like' access to its systems

Analysis

Why This Matters

The arrangement gives an outside body a direct view of a frontier AI lab's internal systems, testing how meaningful independent oversight can be in practice.
It comes amid rising concern that more capable systems could cause harm at scale, and other labs have proposed third-party evaluation as a key safeguard.
The outcome could set a precedent for how the broader AI industry is supervised.

Background

Anthropic is the company behind the Claude family of AI models. METR is a nonprofit research organization that evaluates the capabilities and risks of advanced AI systems and has worked with several leading labs. The push for third-party evaluation reflects a widening debate about how to keep increasingly powerful AI under control.

Key Perspectives

Anthropic: The company has committed to increasing safeguards, citing concerns that its technology could run catastrophically out of control.

METR: The nonprofit's team now works inside Anthropic's office with access to hardware and internal systems, giving it an unusually close view of model development.

Critics/Skeptics: Researchers and policymakers warn that more capable systems could threaten the economy, cyber security and humanity at large, and will judge whether employee-like access translates into real influence over how models are built and released.

What to Watch

Whether METR's findings are published and what they show about the safety of Anthropic's models.
Whether other major labs grant evaluators similar access.
Any public disagreement between the evaluators and the company over risk levels.

Checked

5 claims in this story checked against 1 source: 5 supported.

Show each claim

SupportedAnthropic gives outside AI watchdog 'employee-like' access to its systemsArs Technica - All content: “Anthropic has granted external evaluators “employee-like access” to assess the risks of its models”
SupportedNonprofit METR embeds team in Claude maker's office to assess model risksArs Technica - All content: “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.”
SupportedAnthropic has granted external evaluators “employee-like access” to assess the risks of its AI models, following through on a commitment to increase safeguards over concerns the technology could run catastrophically out of control.Ars Technica - All content: “Anthropic has granted external evaluators “employee-like access” to assess the risks of its models, following through on a commitment to increase safeguards over concerns the technology could run catastrophically out of control.”
SupportedMETR, 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.Ars Technica - All content: “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.”
SupportedA greater role for third-party evaluators is among the main concrete steps the labs have proposed in response to rising concern about the dangers more capable AI systems could pose to the economy, cyber security and humanity at large.Ars Technica - All content: “A greater role for third-party evaluators is among the main concrete steps the labs have proposed in response to rising concern about the dangers more capable AI systems could pose to the economy, cyber security and humanity at large.”