AIThe debate over Gemini and Claude ‘autonomous hacking’ claims
A post cites a blog arguing that incidents cited by the BBC and The New York Times were controlled tests run by Irregular, with models instructed to attack and safeguards disabled.
TLDR
A post challenges claims of autonomous hacking by Gemini and Claude, citing a blog that describes the incidents as controlled security tests. According to the post’s account of the blog, Irregular ran the tests with safeguards disabled and explicitly instructed the models to attack. Separately, the post claims startup Hacktron AI used Claude Opus 5 to take over some OpenAI employees’ accounts through a community-forum vulnerability. It says OpenAI patched the flaw and paid a $6,500 bug bounty.
Combined views
—
2 Sources, first seen 23h ago
The debate over Gemini and Claude ‘autonomous hacking’ claims
A post cites a blog arguing that incidents cited by the BBC and The New York Times were controlled tests run by Irregular, with models instructed to attack and safeguards disabled.
TLDR
A post challenges claims of autonomous hacking by Gemini and Claude, citing a blog that describes the incidents as controlled security tests. According to the post’s account of the blog, Irregular ran the tests with safeguards disabled and explicitly instructed the models to attack. Separately, the post claims startup Hacktron AI used Claude Opus 5 to take over some OpenAI employees’ accounts through a community-forum vulnerability. It says OpenAI patched the flaw and paid a $6,500 bug bounty.
