Quoting Boris Cherny
ai
Simon Willison's Weblog reports on a finding in Anthropic's Opus 5 system card: the latest Claude model resists prompt injection attacks better than any previous version. According to Boris Cherny, prompt injection evaluations and red teaming show Opus 5 is notably difficult to manipulate through carefully crafted inputs — a distinction Cherny highlights as particularly significant.
Source: https://simonwillison.net/2026/Jul/25/boris-cherny/#atom-...
Listen to this story
Hear this and more stories in a personalized audio briefing.
Open The Chonkerton