METR audit says Anthropic may have crossed its own R&D acceleration warning line
What happened
On November 6, The Decoder reported that a METR audit found Anthropic may have crossed its own AI R&D acceleration warning threshold as early as July 15, and that the public risk report understates the pace of acceleration and the uncertainty around it. Since October 2025, AECI test scores for Anthropic models have grown about 1.9 times faster than before, the audit says. METR estimates about 80% of the acceleration comes from internal research use of AI, but the true multiplier may fall between 1.3 and 3, and whether the threshold was crossed is still stated as "may."
Written by AI from the coverage · updated 2 hours ago
Coverage
Follow the reports to see the story from different sides.
- The DecoderMETR finds Anthropic may have crossed its own AI acceleration warning threshold back in July
METR 审计认为，Anthropic 可能早在 7 月 15 日就越过自身 AI 研发加速预警阈值，其公开风险报告低估了加速程度及不确定性。自 2025 年 10 月以来，Anthropic 模型的 AECI 测试分数增长速度约为此前的 1.9 倍，审计方估计约 80% 的加速来自内部研究使用 AI，但实际加速倍数可能介于 1.3 至 3。
Heat over time
Not enough continuous observations to draw a trend yet.