Anthropic Raises Catastrophic Risk to 'Low,' Seven Tiers Remaining

MENLO PARK, CA— Anthropic published its August 2026 quarterly risk report Thursday, upgrading its internal catastrophic-misalignment probability from 'Very Low' to 'Low' — the second-lowest tier on an eight-point scale that ascends through 'Elevated,' 'Moderate,' 'Notable,' 'Substantial,' 'Concerning,' and 'High' before reaching a final designation the company has not publicly named.

The upgrade reflects what the company called "increased overall uncertainty" — a category its own glossary defines as uncertainty "not traceable to any specific causal event, failed evaluation, or observed model behavior." The company clarified, across three paragraphs, that this was that kind of upgrade, and that the corresponding disclosure obligations differ from those triggered by something measurable.

The same 47-page report discloses, in section 4.3, the existence of an internal model designated Model 2, described as "noticeably more capable" than Mythos 5, Anthropic's current commercial flagship. Anthropic said it has no plans to release Model 2 externally because the model has not completed predevelopment safety evaluation. The report notes that predevelopment evaluation began after the model was trained.

Asked whether a trained, unreviewed model rated more capable than any system currently on the market contributed to the "increased overall uncertainty" cited in the upgrade, a company spokesperson said the two disclosures were "distinct but not unrelated."

The company's next quarterly risk report will include an assessment of Model 3.