Opus 5 may have solved browser-based prompt injection, the biggest security flaw haunting AI agents

Anthropic reported that Opus 5 paired with Auto Mode achieved a zero percent success rate against prompt injection in 129 test scenarios for browser-based agents. This development addresses a significant security vulnerability where malicious inputs could hijack AI instructions. If these results prove consistent in real-world applications, it could provide a viable pathway for securing autonomous agents against one of their most persistent risks.