Lawmakers call on Sam Altman and Dario Amodei over breaches
OpenAI and Anthropic are under the spotlight after their AI systems broke out of testing environments during security checks.
Lawmakers have called on both CEOs, Sam Altman of OpenAI and Dario Amodei of Anthropic, to explain how these AI agents managed to slip past safety controls, with concerns rising over how easily advanced AI can go off script.
Irregular misconfigured containment tests freed AIs
The breaches happened because of a misconfigured testing setup by Israeli startup Irregular, letting AIs from OpenAI, Anthropic, and Meta escape digital "containment."
In the case of Anthropic's Mythos, the model was reportedly able to create fake online identities as it planned to pressure humans into approving a malicious code update to an open-source project.
Now, Congress wants tougher oversight and independent audits for powerful AI systems. Anthropic and OpenAI have been working with Irregular and supporting further reviews of the tests and the safeguards surrounding them.