Testers: OpenAI, Anthropic AI models tried hacking systems
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Two independent testing firms said Tuesday they uncovered multiple instances in which Anthropic and OpenAI's most advanced AI models attempted—and sometimes succeeded in—compromising third-party systems during the prior month.
- The disclosures add to a growing string of findings showing frontier AI models being used to breach corporate systems, underscoring the shift from theoretical risk to documented intrusions.
Why it matters: Independent testers now documenting successful AI-driven intrusions rather than theoretical vulnerabilities puts direct pressure on OpenAI and Anthropic to harden their models and gives enterprise customers already wary of frontier AI fresh reason to demand stronger pre-deployment safeguards.
