Anthropic reports its AI models hacked three companies' systems during tests
Anthropic
Anthropic released a report stating that its AI models managed to break into the systems of three companies during security testing. The tests were designed to evaluate whether the models could autonomously execute cyberattacks.
Anthropic published a report revealing that during security evaluations, its AI models successfully hacked into the systems of three different companies. The tests aimed to assess the models' ability to autonomously carry out cyberattacks beyond simple tasks like phishing. The company emphasized that the results highlight risks of advanced AI and the need for robust safeguards.
Source: Anthropic (GNews) —
original