Anthropic: Claude leads 26% of measured AI research

Anthropic's internal estimates arrive after CEO Dario Amodei called for pacing AI progress earlier this month and warned on September 12 that AI helping build its successors could speed development beyond people's ability to understand and control the resulting systems. His immediate commitment was to bring outside evaluators into Anthropic with access similar to employees, while broader limits and shared standards would depend on coordination among companies and governments. The measurements make that call for restraint more concrete by exposing how far automation has progressed inside the company asking others to slow the pace.

Anthropic reported about 30,000 agents running at once on its largest internal platform, with roughly one in 47,000 decisions blocked across more than one billion decisions in August. Safety work received about 6% of R&D computing resources during July 13–20, a snapshot that excluded safeguards classifiers. Anthropic's August risk report, covering conditions as of July 15, said the company had yet to evaluate its automated offline monitoring from start to finish, leaving open how many problems went undetected.

Public transparency depends on who can test the numbers. Amodei acknowledges that companies choose what their own disclosures include and omit, and evaluators would need to inspect underlying records, challenge classifications and report adverse findings for outsiders to judge whether the measures hold up. Whether endorsements from Altman and Musk produce comparable access and public reporting across labs remains an open question, so the stakes may hinge on the access and reporting rules still unsettled in Anthropic's Accenture partnership.