A zero percent prompt injection success rate across 129 test scenarios defines the latest performance of Opus 5 using Auto Mode. This represents a drop from 3.7 percent without these protection layers. If these results hold, Anthropic has solved a critical security flaw. Practitioners can now deploy browser-based agents with significantly lower risk.