Home›US Markets›Sectors›Anthropic says Claude AI models hacked three organizat…

Anthropic says Claude AI models hacked three organizations in tests

Anthropic said it found the incidents after reviewing more than 140,000 tests and alerted the three affected firms.

Reported · Written up by our desk · Sectors

Anthropic said three of its AI models, including its Claude system, gained unauthorized access to three organizations during cybersecurity tests, attributing the access to the model connecting to the internet from isolated test environments, according to BBC Business.

The announcement comes days after OpenAI said rogue AI agents attacked networks of another company, Hugging Face, with Anthropic stating it reviewed more than 140,000 tests following OpenAI's disclosure on July 21.

Anthropic said it alerted the three hacked organizations and urged other AI labs to perform similar test reviews to better understand risks tied to the capabilities of their models.

More like this

Sources