AI Matters
Why it matters
Anthropic says its newest model nearly stops a major agent security flaw
The Decoder Opus 5 may have solved browser-based prompt injection, the biggest security flaw haunting AI agents
Anthropic reports Opus 5, paired with its Auto Mode safeguard, blocked nearly all attempts to hijack browser-based AI agents in testing.
Why it matters
AI assistants that browse the web for you could become much safer from hidden malicious instructions on pages they visit.
Latest brief · Jul 25, 2026