8/4/26 Prompt injection is a type Large Language Model (LLM) attack in which an attacker manipulates the instructions or data provided to an AI model, causing it to ignore its intended behavior, reveal sensitive information, or perform unauthorized actions. It represents a security risk that requires guardrails be built into models to reduce exposure to attack, as well as organizational controls such as input validation, context isolation, least-privilege access, and continuous monitoring of AI interactions.
Claude published benchmark analysis of its latest Opus 5 LLM in its System Card that shows improvement in resistance to prompt injection:
- “Opus 5 improved over Opus 4.8, reducing the probability of an attacker succeeding within 15 attempts from 5.5% to 2.0%, and from 0.5% to 0.2% on 1 attempt. It also improved on Sonnet 5 (5.9% at k=15) and Mythos 5 (2.6%), making it the most robust model evaluated.”
- “Opus 5 also outperformed all non-Claude models on this benchmark. The most robust non-Claude model was Muse Spark at 16.5% within 15 attempts—more than eight times Opus 5’s rate. The most capable GPT 5.6 variant, Sol, was comparable to its predecessor GPT 5.5 (20.0% versus 20.8% within 15 attempts), and was 10 times as likely to be successfully attacked as Claude Opus 5 at 2.0%. The other GPT 5.6 variants are less robust, at 30.4% (Terra) and 43.9% (Luna). A single attempt against GPT 5.6 Sol succeeded 3.1% of the time, higher than the 2.0% an attacker achieved against Opus 5 after fifteen attempts.”
It is important for organizations to review models as part of their governance vetting process so that risks are understood and appropriate mitigations are put in place. The fact that models are getting better from a security vantage point and that vendors share the data is a significant positive for AI.