Claude Opus 5 system card published

Alongside its launch of Claude Opus 5, Anthropic published the accompanying system card, its standard pre-deployment evaluation document covering capability, safety and welfare testing. The company reported Opus 5 showed no new concerning alignment properties relative to prior models and assessed its overall alignment risk as very low. On Anthropic’s automated behavioural audit, Opus 5 scored above Sonnet 5, Opus 4.8 and Mythos 5, with the company describing it as its most aligned model to date and citing particularly strong adherence to Claude’s constitution, alongside rising alignment scores and falling rates of cooperation with misuse and reckless behaviour requests.

On capability thresholds under Anthropic’s Responsible Scaling Policy, the company said Opus 5 did not cross the automated AI R&D capability threshold that would trigger additional safeguards, and classified it at the CB-1 tier for biological and chemical weapons assistance — able to help with existing, non-novel threats but not assessed as enabling genuinely novel ones.

The card also included a model-welfare section, a practice Anthropic began treating as routine through 2026. It reported Opus 5 showed the highest and most consistent self-rated sentiment of any model the company had evaluated, with affect assessed as neutral to mildly positive across training, deployment and behavioural audits, based on automated interviews, behavioural observation and analysis of the model’s self-reported preferences.

As with prior system cards, the findings rested on Anthropic’s own evaluation framework and were not independently verified, though the document’s publication alongside the launch — rather than after it — continued the practice GPT-4’s system card had established of treating dangerous-capability and welfare evaluation as a pre-deployment step rather than an afterthought.