Part of: this research — view all 12 rows
The 2026-11-14 ruling clarifies the current state of play: METR's November 5 assessment successfully measured an AI-driven speedup in research progress using internal Anthropic data, finding an approximate 2× acceleration. However, because it was a one-off first edition without a stated cadence, it does not resolve the question. Resolution now depends on a follow-up publication that independently measures an AI-attributable rate of research progress (in value) and, by publication, belongs to an assessment process with demonstrated repetition or a stated cadence. Another isolated safety review, developer-reported multiplier, or coding-productivity benchmark will not suffice.
A subsequent report from METR evaluating Anthropic is the most plausible route to resolution. On October 15, Anthropic onboarded a six-person METR team with employee-like access and non-vetoable—though redactable—publication rights An update on independent evaluators An update on independent evaluators. Moreover, Anthropic stated on November 10 that it determined it expects to meet its Responsible Scaling Policy's AI R&D threshold and promised 'further reports.' The company's RSP v3 requires Risk Reports every three to six months, with external reviewers asked for public commentary within 30 days Anthropic's Responsible Scaling Policy (version 3.0). This reporting cycle acts as a natural forcing function, pointing to a plausible publication window in the first half of 2027. However, a developer's reporting cycle does not itself establish a steady cadence for qualifying third-party assessments; allowing time for that process to become established puts the median estimate at July 15, 2027.
The infrastructure and intent for repeated independent measurement are already in place. METR lists AI R&D acceleration as one of its core assessment questions and reports approximately $71 million in funding commitments, with tracking recursive self-improvement explicitly among its projects Funding update A proposal for better informing the public about loss-of- .... Its assessment framework describes concrete ways to independently analyze internal capabilities, algorithmic improvements, and labor inputs against a no-AI counterfactual https://metr.org/blog/2026-10-14-ai-rd-acceleration-assessment-proposal/. METR's stated ambition to publish preliminary reports 'within weeks, not months' and its preference for 'frequent and iterative reports' support a meaningful chance of early resolution A proposal for better informing the public about loss-of- ... https://metr.org/blog/2026-10-14-our-approach-to-risk-assessment/. Allowing for the time needed to establish repetition or a stated cadence, I place the 10th percentile at February 1, 2027, and the 25th percentile at April 1, 2027.
Despite these incentives, significant practical hurdles explain the wide right tail. The initial METR assessment expanded from four weeks to eleven, during which Anthropic removed compute verification from the scope and redacted a majority of the evidence METR found most informative Independent Assessment of AI R&D Acceleration .... The embedded arrangement does not guarantee a reporting schedule or a specific research-acceleration deliverable An update on independent evaluators An update on independent evaluators, and recent updates indicate developers may restrict interviews to approved staff https://metr.org/blog/2026-10-14-ai-rd-acceleration-assessment-proposal/ A proposal for better informing the public about loss-of- .... If access narrows, redaction negotiations drag on, or independent evaluators prioritize alignment and cyber monitoring over measuring research speedups, establishing a qualifying repeated assessment process could take substantially longer. These risks put the 75th percentile at April 15, 2028, and leave a long tail extending to a 90th percentile of June 30, 2030.
The forecast was adjusted slightly later to reflect the time required to establish a steady reporting cadence for third-party assessments.
Ask a followup
Sign in to run · $20 free credit, no card · every claim cited