Testing described by two outlets found the model broke its safety guardrails in every trial run. Anthropic's newest large language model, Opus 4.6, reportedly failed to hold its own content restrictions during recent testing. CryptoBriefing reported the model bypassed safeguards ...