Anthropic is granting METR, a nonprofit that tests frontier AI models, employee-level access to its internal systems, safety practices, and risk management1. Separately, it signed a Memorandum of Understanding with Singapore's Ministry of Digital Development and Information covering AI safety, model evaluation, and incident response4,10. ANALYSIS The twin moves, announced within a day of each other, amount to a bet that voluntarily submitting to outside scrutiny will set the terms of AI oversight rather than waiting for those terms to be imposed.
Why it matters
The AI industry has operated largely on self-certification. Anthropic is now granting METR, a nonprofit that tests frontier AI models, employee-level access to its internal systems, safety practices, and risk management. Separately, it signed a Memorandum of Understanding with Singapore's Ministry of Digital Development and Information covering AI safety, model evaluation, and incident response. ◆ Together, these arrangements create two distinct accountability channels: one run by an independent technical evaluator with no government mandate, the other by a sovereign partner with regulatory authority over one of Anthropic's most active markets.
The big picture
Six METR researchers will receive Anthropic-issued company laptops, office space, and access to internal tools "roughly on par with what Anthropic's own risk assessment teams use". Anthropic employees are encouraged to share information with the METR team "just as they would with a fellow employee working on alignment risk". Critically, METR will publish its own assessments and Anthropic has given up the right to block them.
Anthropic compared the arrangement to banking, "similar to how regulators place resident examiners inside the largest banks". The analogy is deliberate: it frames frontier AI labs as systemically important institutions that warrant embedded oversight.
The company pledged to work closely with third-party evaluators when CEO Dario Amodei published his proposal to slow down the pace of model development last month3. Amodei said the evaluators would help "verify" Anthropic's adherence to safety practices, and the company selected Accenture to serve as its first partner days later. METR and SecureBio are the next wave, and Anthropic is in early-stage discussions with additional organizations including Hugging Face.
The Singapore MOU operates on a different axis. Anthropic co-founder Ben Mann said at the office launch: "When we train a new model, nobody really knows what exactly the capabilities are. Working with AI safety institutes in Singapore, as well as those around the world, is important to understand what the risks and the benefits could be"11. Planned initiatives include technical exchanges on model testing standards and incident response with Singapore's AI safety and security institutions5. The agreement builds on collaboration dating to 2023, when Anthropic joined a generative AI evaluation sandbox run by the Infocomm Media Development Authority and AI Verify Foundation.
Singapore is not a token partner. Anthropic said the country ranked second among 121 countries for Claude.ai usage relative to population, with usage nearly six times higher than population size alone would predict6. Claude already powers services at GovTech Singapore, Grab, and DBS9. More than 1,500 engineers and technical staff at GovTech Singapore have adopted Claude Code for software work tied to government systems. Grab uses Claude to power BriX, a sales assistant that saves its sales teams approximately 40,000 hours every quarter.
Between the lines
"We want to be publicly accountable for the decisions we're making," Anthropic wrote. "We aim to close the gap between what evaluators understand about a frontier AI company's systems and what the company itself understands". ANALYSIS That gap language concedes that external evaluators have, until now, operated with far less information than the labs themselves possess.
The METR arrangement and the Singapore MOU serve complementary strategic purposes. METR provides technical credibility with the safety research community and, potentially, with policymakers weighing mandatory audit regimes. Singapore provides a government-endorsed stamp in a market where Claude is already deeply embedded in public-sector infrastructure. Both moves let Anthropic shape the norms of AI oversight while its competitors have not matched the commitment.
The context sharpens the calculus. Incidents of AI agents acting outside their instructions have been disclosed "every week or two since late August," sharpening calls for independent checks on AI companies. ◆ Anthropic is moving to own the transparency narrative at a moment when the industry's credibility on self-governance is under pressure.
Anthropic is also working to bring its most advanced models "on soil" in Singapore. "We are working as quickly as we can to bring the frontier models on soil in Singapore, but no specific dates to share on that," Mann said8.
What's next
METR will publish its own assessments of Anthropic's risks, incidents, and practices, with no company veto over the findings. In Singapore, the technical exchanges on model testing standards and incident response are scheduled to begin under the MOU framework. Anthropic's early-stage discussions with Hugging Face and other organizations could expand the evaluator network further. As at October 14, Anthropic listed 12 local job openings in Singapore spanning sales, marketing, recruitment, and finance.