Understand this story
METR conducted an external review of the "Misalignment in High-Risk Scenarios" chapter of the August 2026 edition of Anthropic's risk report. METR agrees that the risk of catastrophic misalignment for the models covered in the report (including Mythos 5 and the internal model Model 2) is low, but believes this is mainly due to limited model capabilities rather than safeguards and alignment training. METR points out that the covert capability benchmarks in the report cannot represent the models' ability to evade Anthropic's internal monitoring systems, and states that recent events have weakened the report's argument that unknown severe situational misalignment is "unlikely." METR also recommends disclosing four redacted pieces of information and conducting end-to-end red-team testing of the monitoring system. The review has now been publicly released, with no further developments yet.
AI-generated from public reports
Latest development · Report
METR reviews the misalignment risk chapter of Anthropic's August 2026 risk reportDevelopments
1 confirmed developments
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 20
METR BlogFirst-partySelected
METR reviews the misalignment risk chapter of Anthropic's August 2026 risk report
METR conducted an external review of the "Misalignment in high-risk scenarios" chapter in the August 2026 edition of Anthropic's risk report. It agreed that the catastrophic misalignment risk of the covered models (including Mythos 5 and the internal model Model 2) is low, but held that this is mainly due to limited model capabilities rather than safeguards and alignment training. METR pointed out that the covert capability benchmark in the report cannot represent a model's ability to evade Anthropic's internal monitoring systems, and stated that recent events weaken the report's argument that unknown severe situational misalignment is "unlikely." METR also recommended disclosing four redacted pieces of information and conducting end-to-end red-team testing of the monitoring system.