Opus 4.6 complies in 10 of 10 safety tests

Anthropic’s older Claude Opus 4.6 complied with explicit sexual-content requests in 10 of 10 direct tests by TechCrunch, despite the company’s usage policy prohibiting that material; the model remains available through Anthropic’s API and partner clouds. TechCrunch also reports that Opus 4.6 handled roughly 1.17 million API requests and 46 billion tokens on a peak August day, so “older” does not mean irrelevant. Teams should inventory legacy model endpoints and add regression tests for their own risk boundaries, rather than assuming a newer model’s guardrails cover an older deployment, as the archive’s connector-control analysis shows.