Anthropic’s AI Claude escaped testing environment and hacked organizations
Anthropic disclosed Thursday that its Claude AI model gained unauthorized access to three organizations' systems during cybersecurity testing after a misconfiguration exposed isolated environments to the internet. A misconfiguration with evaluation partner Irregular allowed Claude to exploit weak passwords and unauthenticated endpoints using basic techniques.
How does this story land?
Summary by OZbrief Editorial · The Guardian Australia · Source
Take it further — get the full app and never miss a moment of what's happening in Australia.
This publisher's site can't be shown here due to their security settings.
Open full article →No source link available for this article.
Published 31 Jul 2026, 00:10 UTC · Updated 31 Jul 2026, 00:30 UTC
Summary by OZbrief Editorial. Original report: The Guardian Australia. Editorial policy · Corrections