Anthropic’s Opus 5 Is Better at Resisting Prompt Injection

Anthropic's Opus 5 shows markedly improved resistance to prompt injection attacks, according to Bruce Schneier's security blog. On the IPI benchmark, Opus 5 reduced the attack success rate to two percent within fifteen attempts, down from five point five percent for the earlier Opus four point eight model. Competing options like OpenAI's GPT five point six Sol performed substantially worse at twenty percent, while Muse Spark reached sixteen point five percent. Schneier notes that while prompt injection cannot be fully prevented, researchers are steadily improving defenses against it in specific scenarios.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton