Claude Opus 5 is Here: What Developers Need to Know About the Safety "Fine Print"

Anthropic has released Claude Opus 5, with major gains in alignment, coding, and prompt-injection resistance.

The model is strong at finding software bugs but still limited in producing full exploits, and it remains at ASL-3.

Anthropic now allows defensive source-code vulnerability finding, while continuing to block vulnerability hunting against compiled binaries.

The system card shows a notable safety gap: Claude.ai has extra system-level safeguards, while API users must add their own guardrails.