Quoting Boris Cherny
https://twitter.com/bcherny/status/2080713091688583312">More than any of these eval scores, what is most exciting to me is something else: Opus 5 is our least prompt injectable model yet.
It is a bit buried in the system card, but across PI evals and red teaming, Opus 5 is very hard to prompt inject successfully.
ā https://twitter.com/bcherny/status/2080713091688583312">Boris Cherny, here's that https://www-cdn.anthropic.com/c5fbac3f0b1280a933ebd26d3cb8bb9f5bdeaf48/Claude%20Opus%205%20System%20Card.pdf#page=73">System Card section, page 73
Tags: https://simonwillison.net/tags/prompt-injection">prompt-injection, https://simonwillison.net/tags/anthropic">anthropic, https://simonwillison.net/tags/claude">claude, https://simonwillison.net/tags/generative-ai">generative-ai, https://simonwillison.net/tags/ai">ai, https://simonwillison.net/tags/llms">llms, https://simonwillison.net/tags/boris-cherny">boris-cherny