Anthropic Doubles Its AI Speed While Publicly Calling for a Slowdown

Anthropic, the San Francisco–based company behind the Claude large language model, has been found to be developing its AI systems at roughly twice the pace it maintained a year ago, according to a new analysis by the nonprofit Model Evaluation and Threat Research (METR) based in Berkeley, California. Anthropic granted METR access to internal data and staff interviews, making this one of the more detailed third‑party looks into a leading AI lab’s development pipeline.

The finding is striking because Anthropic has positioned itself as a cautionary voice on AI risk, frequently urging tighter regulation and even a slowdown of frontier‑model development. Internally, however, the company appears to be shortening iteration cycles and aggressively scaling model capabilities and compute, partly as it prepares for a highly anticipated IPO that will require convincing public markets of its growth story.

The report highlights the widening gap between public rhetoric and internal incentives across the AI industry. As competitive pressure from rivals intensifies, safety‑minded labs face strong financial and strategic reasons to move faster, not slower. Regulators and civil‑society groups are likely to seize on these findings as fresh evidence that external audits, mandatory risk disclosures, and clearer rules on development pace may be needed to keep frontier AI efforts aligned with their stated safety commitments.