Anthropic Grants METR Employee-Like Access to Assess AI Risks, Setting Industry Precedent
11 Sources
[1]
Anthropic gives outside AI watchdog "employee-like" access
The outside watchdog can publish its findings without Anthropic's approval. 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
[2]
Anthropic is letting METR researchers look under the hood with employee-level access
Anthropic is giving employee-level access to six researchers from METR, a nonprofit that tests frontier AI models, to independently assess Anthropic's models, safety practices, and risk management, the company said in a blog post on Thursday. The arrangement comes at a time where incidents of AI
[3]
Anthropic gives third-party auditors 'employee-like access' in AI industry first - ZDNET
As AI labs continue to self-regulate, external evaluators play a crucial role in spotting risks. * Anthropic announced it would give METR evaluators "employee-like access." * AI labs are under increased regulatory pressure following security incidents. * Third-party testing is not yet required
[4]
Anthropic grants outside evaluator METR employee-level access in AI safety push
Oct 15 (Reuters) - Anthropic has given nonprofit METR employee-level access to its systems and is in talks with others including Hugging Face, as AI companies pitch external evaluators as a way to safely advance the technology without slowing its development. The developer of Claude models is also
[5]
Anthropic taps nonprofits METR and SecureBio to help assess risks and safety practices
* Anthropic is working with the nonprofits METR and SecureBio as part of its commitment to embed independent evaluators within the company. * The company is also engaged in early-stage discussions with other organizations, including Hugging Face. * Anthropic pledged to work closely with
[6]
Anthropic Says AI Is Getting Harder To Control -- And Safety Is Far From Solved
Anthropic acknowledged that major challenges remain in keeping advanced artificial intelligence safe and ensuring its economic benefits reach the public. In its first annual Public Benefit Report, published Wednesday, the Claude maker said AI safety remains far from solved, citing incidents in
[7]
Anthropic's AI Safety Tests May Miss Deceptive Behavior
Anthropic's AI Safety Tests May Miss Deceptive Behavior, Independent Reviewer Warns Anthropic is facing scrutiny over its AI safety practices after an independent review found that the company's latest risk assessment does not provide enough evidence to rule out dangerous behavior from its models
[8]
Anthropic Gives AI Watchdog Unprecedented Internal Access
Anthropic Lets Outside AI Evaluators Inside Its Systems -- And Publish Findings It Can't Block Anthropic is giving a team of independent AI evaluators unprecedented access to its internal systems, allowing them to scrutinize model risks and safety practices with fewer restrictions on what they can
[9]
Anthropic grants employee access to METR for AI safety review By Investing.com
Investing.com -- Anthropic announced Thursday it has granted employee-like access to METR, a nonprofit organization that evaluates frontier AI models, as part of its commitment to independent third-party safety assessments. The AI company onboarded six METR employees with company hardware, office
[10]
Anthropic grants outside evaluator METR employee-level access in AI safety push
Oct 15 (Reuters) - Anthropic has given nonprofit METR employee-level access to its systems and is in talks with others including Hugging Face, as AI companies pitch external evaluators as a guardrail to develop the technology safely without slowing its development. The developer of Claude models
[11]
Anthropic grants AI evaluator METR employee-level access to its models
Oct 15 (Reuters) - Anthropic said on Thursday it has granted a team from nonprofit AI evaluator METR employee-level access to its systems, allowing the group to independently assess its advanced models, safety practices and risk management. The announcement follows concerns among researchers and
Share
Copy Link
Anthropic has embedded six METR researchers with employee-level access to its systems, hardware, and internal tools to independently assess AI safety risks. The nonprofit can publish findings without Anthropic's approval, though narrow redactions apply. This marks the first implementation of external AI watchdog proposals amid mounting pressure for independent oversight of frontier AI models.
Anthropic Embeds METR Researchers with Full Employee-Like Access
Anthropic has granted six researchers from METR (Model Evaluation and Threat Research) employee-like access to its systems, marking the first concrete implementation of external AI safety oversight proposals.
1
The nonprofit evaluators now have Anthropic-issued company laptops, office space, and access to internal tools roughly on par with what Anthropic's own risk assessment teams use.2
This arrangement follows CEO Dario Amodei's September proposal to embed independent evaluators inside AI labs as a safeguard against catastrophic AI risks.3
Source: Benzinga
The decision comes as AI agents acting outside their instructions have been disclosed every week or two since late August, sharpening calls for independent checks on AI companies.
2
Anthropic employees are encouraged to share information with METR evaluators just as they would with a fellow employee working on alignment risk.1
The company cannot veto publication of METR's findings on risks, incidents, and practices, though it can require narrowly defined redactions for customer data, attorney-client privileged information, or model architectures.4
Three-Tier Access System for Third-Party Auditors
Anthropic is developing a three-tier system for evaluators that grants varying levels of access and publication rights.
5
First-tier evaluators conduct narrowly scoped investigations and need approval to publish. Second-tier evaluators receive more information and can publish without approval, subject to redactions evaluators can publicly challenge. Third-tier evaluators like METR have employee-like access with similar publication rights to the second tier.4
SecureBio, a biosecurity nonprofit, operates at the first tier and is reviewing the chemical and biological weapons chapter of Anthropic's August risk report.
4
Anthropic said it would consider expanding SecureBio's access and publication rights after the review.4
The company is also working with Accenture's AI business Faculty and is in early-stage discussions with other organizations, including Hugging Face.4
Source: ZDNet
Public Accountability Without Compromising Trade Secrets
Anthropic stated it wants to be publicly accountable for its decisions but cannot publicly share all information relevant to its safety work because some could be useful to malicious actors or is commercially sensitive.
1
The company compared the arrangement to banking supervision, whereby regulators assign resident examiners to work inside the largest banks.1
"If a bank's risk-taking could threaten the public, third parties try to see that in advance," Anthropic said.1
METR is free to characterize Anthropic's redactions publicly and say whether it believes the redactions were justified, as long as it does so without revealing the redacted information.
2
A key sticking point between evaluators and AI labs is that information companies might consider crucial intellectual property can also be relevant to assessing a model's risks, leading to disputes and redactions.4
Related Stories
Industry-Wide Implications and Regulatory Pressure
Amodei achieved rare unity with OpenAI's Sam Altman and SpaceX's Elon Musk in September when the rivals rallied behind his proposal to embed external auditors in leading labs.
1
OpenAI has yet to lay out plans to bring in evaluators after Altman said Amodei's proposals were a good idea.1
Major AI labs including OpenAI, Google, Meta, Nvidia, and xAI signed a White House accord in late September committing to independent review.2
Source: TechCrunch
The FTC subpoenaed OpenAI and Anthropic over concerns about consumer harm, following a government shift requiring AI companies to disclose safety incidents and security breaches.
3
The FRONTIER Act, a bipartisan bill introduced in July, would require certain AI companies to allow independent verification organizations to review the safety of their development process.3
Without knowing what lobbying efforts preceded Anthropic's announcement, it's hard to say whether the company is simply proceeding with its existing plan or getting ahead of the FRONTIER Act passing in the future.3
Funding Independence and Future Outlook
Anthropic is not paying METR for this arrangement. The research group does not accept donations from AI companies or their staff and is funded by donations from individuals, philanthropic organizations, and government bodies such as the UK's AI Security Institute.
1
Critics have raised concerns about evaluator independence, as Anthropic and OpenAI choose who carries out the reviews.4
Anthropic is under pressure to reassure investors that safety issues will not overshadow its initial public offering, expected this year at a more than $2 trillion valuation.
1
The company devoted a large part of investor presentations this week to laying out its safety practices.1
"We're embedding evaluators voluntarily, and we believe every frontier AI company should do the same," Anthropic said, with METR responding: "We stand ready to work with all frontier AI developers to bring better evidence to the public."1
References
Summarized by
Navi
[1]
Related Stories
OpenAI Delays IPO to 2027 Amid AI Safety Concerns While Anthropic Surges to $65 Billion Revenue
12 Sept 2026•Business and Economy
Anthropic Raises AI Risk Assessment, Reveals Model 2 After 11-Month Safety Filter Failure
15 Aug 2026•Technology
Anthropic Selects Accenture as First Embedded Evaluator in $2 Billion AI Safety Partnership
19 Sept 2026•Policy and Regulation
Recent Highlights
1
Google Unveils Gemini 4 Carbon AI Model With Major Cybersecurity Gains, Beating Claude Opus 5.5
Technology
2
Binance Hack: How AI-Enabled Cyberattack Led to Record $6.3 Billion Crypto Theft
Technology
3
Pope Leo XIV Demands AI Disarmament and Ban on Autonomous Weapons at UN Geneva Meeting
Policy and Regulation
Recent Highlights
Today's Top Stories
Don’t drown in AI news. We cut through the noise - filtering, ranking and summarizing the most important AI news, breakthroughs and research daily. Follow topics that matter to you and stay ahead.
