METR reviews the misalignment section of Anthropic’s August risk report
METR’s external review of the “Misalignment in high-stakes settings” section of Anthropic’s August 2026 risk report agrees that the risk of catastrophic harm from misalignment is low, but mainly because of models’ limited capabilities rather than safeguards and alignment training; it says recent incidents undermine some of the report’s arguments.
- Published
- Added to the archive
- Source checked on
- Original title
- Review of the "Misalignment in high-stakes settings" section in Anthropic's Risk Report (August 2026)
Details
METR says Anthropic’s July 24 draft rated the risk “very low” and was revised to “low” after a group of incidents involving its models came to light, with most content unchanged; METR considers the incidents observed by UK AISI better described as “unknown severe context-dependent misalignment,” which the report calls “unlikely.” Citing Anthropic, METR notes the offline monitor flagged only about 1% of actions in the Mythos 5 incident (about 50% with the chain of thought removed), and it judges two of the five informative redactions it examined not properly justified. The review is lightweight and based solely on the internal report; Anthropic could veto or redact it, and METR says no redactions were made and it changed only wording and one experiment’s method in response to feedback.