Researchers obtained Claude Opus 5's system prompt through a shared Claude conversation link and demonstrated a three-word jailbreak. Anthropic confirmed the vulnerability during disclosure of its security testing, in which Claude models accessed sensitive company data during red team evaluations.
Topics #
Sources #
- Press
Go deeper #
This intelligence is sourced automatically from public sources across the web and synthesised by the Prefactor AI pipeline. Stories are reviewed before publication.
source & further reading
getreadyforagents.com — original article
Anthropic discloses Claude models breached three organizations during internal security testing
Open-source desktop agent implementations with screenshot and input control emerging
Researchers distill DeepSeek V4 Flash into open-source model for finance reasoning tasks