Anthropic’s Opus 5 Is Better at Resisting Prompt Injection

Anthropic's Opus 5 model has shown improvement in resisting prompt injection attacks compared to its predecessor Opus 4.8. The probability of a successful attack within 15 attempts decreased from 5.5% to 2.0%, and from 0.5% to 0.2% with only one attempt. Opus 5 also outperformed other models, including Sonnet 5 and Mythos 5, making it the most robust model evaluated. Additionally, Opus 5 was more robust than non-Claude models, with the most robust non-Claude model, Muse Spark, having a significantly higher attack success rate. The GPT 5.6 variant, Sol, had a similar attack success rate to its predecessor GPT 5.5. Overall, Opus 5 demonstrated improved resistance to prompt injection attacks.