News · curated 25 Jul 2026

Quoting Boris Cherny

Coverage timeline

25 Jul 2026simonwillison.netprimary

Single-source analysis — first reported, latest, and curated coincide.

Why it matters

Prompt injection robustness of deployed agentic models like Claude Opus 5 directly affects the attack surface defenders must reason about when deploying AI agents and copilots.

A quotation from Anthropic's Boris Cherny, collected by Simon Willison, highlights that Claude Opus 5 is described in its system card as Anthropic's 'least prompt injectable model yet,' with the card reporting the largest agentic-safety gains in prompt injection robustness across coding, computer use, and browser use.