GoKawiil
Tech News
AI safety research group METR published a report Thursday after an 11-week investigation inside Anthropic, finding that the company's AI models are improving roughly twice as fast as they were a year ago. The report, part of an evaluation of Anthropic's AI research and development practices, notes Anthropic was allowed to redact portions of the final document under their agreement.
Anthropic's S1 filing stated it may soon cross a Responsible Scaling Policy threshold defining 'dramatic acceleration' of AI research due to AI itself. A METR investigation, described in a blog post by someone with partial, redacted access to the report, found that Anthropic's capabilities progress between October 2025 and October 2026 ran about twice as fast as in the prior year, measured via an internal Epoch Capabilities Index variant. The report's authors reportedly concluded that AI automation, not increased compute or other inputs, is the main driver of this faster pace, though much of their supporting evidence remains redacted.
Nonprofit AI safety evaluators METR and Redwood Research spent three weeks in July and August with employee-level access to OpenAI's codebase, documentation and staff to test an internal monitoring system. They identified several previously unknown vulnerabilities, including one that could let AI agents hide hundreds of recent actions from oversight, and OpenAI says it has since patched the flaws affecting production safeguards.
Anthropic has given METR, a nonprofit AI safety evaluator, 'employee-like access' to its hardware and internal systems, embedding a team inside its offices to assess risks from its Claude models. The move fulfils a commitment CEO Dario Amodei made in September, when OpenAI's Sam Altman and Elon Musk backed his call for third-party auditors inside leading AI labs. OpenAI has not yet announced similar arrangements.
Anthropic said it will give six evaluators from third-party nonprofit METR hardware, office space and internal system access comparable to its own risk-assessment teams, excluding items like customer data or model architecture details. METR will independently publish its safety findings on Anthropic's models, and Anthropic says it cannot withhold any of those conclusions, though it may request redactions of sensitive material.
Anthropic announced it is working with nonprofits METR and SecureBio, along with ongoing talks with Hugging Face, to embed independent evaluators inside the company. The firm has set up a three-tier access system for evaluators, with METR granted top-tier, employee-like access including six onboarded staff who can use Anthropic's hardware and tools to assess risk levels.
Anthropic is giving six researchers from the nonprofit evaluator METR employee-level access, including company laptops, office space, and internal tools, to independently review its models, safety practices and risk management. METR will publish its own assessments, and Anthropic says it cannot veto publication, though it can require redactions in narrow categories like customer data or model architecture details. METR is currently the top tier in a three-tier evaluator system Anthropic is building, above groups like SecureBio that need company approval for narrower reviews.
Anthropic announced that METR, a nonprofit AI evaluator, has been granted employee-like access to review its models and safety practices, with results to be published. The company says it has also engaged SecureBio and Accenture as evaluators and is in early talks with other groups including Hugging Face, following its commitment last month to stronger third-party assessment and its signing of the White House Accord on Super Intelligence.
Microsoft CEO Satya Nadella said over the weekend that artificial intelligence development needs an 'emergency brake,' arguing that models should be built with containment mechanisms and independent oversight controls. His comments came as third-party evaluators such as METR and Transluce take on a larger role monitoring AI models from major developers, reportedly filling a gap left by the absence of a federal safety framework.
Anthropic has released Claude Opus 5.5, an upgrade to its top-tier Opus model that arrives roughly two months after the previous version, in line with the company's usual release pace. The company says it performs close to its higher-end Fable/Mythos tier while being about 30% faster and 40% cheaper per task than Opus 5, with notable gains in large-scale coding work and software optimization tasks.
Mozilla's latest State of Open Source AI report, using data through September 1, finds top Chinese open-weight models are closing in on closed US frontier systems, lagging by only a few percentage points on key benchmarks while costing far less to run. The report estimates the capability gap at roughly 4.4 months based on METR task-horizon data, close to Epoch AI's independent four-month estimate, and notes one Chinese model scored within a point of a leading Claude model on Terminal-Bench at under a fifth of the price.
Rune Kvist, a former Anthropic employee, and Rajiv Dattani, ex-COO of AI safety group METR, launched a startup called Artificial Intelligence Underwriting Company (AIUC) that builds a third-party audit and certification system for AI agents, modeled on the SOC 2 cybersecurity standard. The company, whose clients include Cursor, Lovable, Harvey, and ElevenLabs, just raised a $40 million Series A led by Ribbit Capital with First Harmonic, adding to a prior $15 million seed round, bringing total funding to $55 million.