Anthropic Probes Unauthorised Access to High-Risk Mythos AI Model

53 articles · Updated · BBC.com · Apr 22

Anthropic is investigating reports that unauthorized users gained access to its powerful Mythos AI cybersecurity tool via a third-party contractor.
The group reportedly used existing access and online sleuthing techniques to enter Mythos, which is designed to identify and exploit software vulnerabilities.
Although the group claims no malicious intent, the incident raises concerns about controlling advanced AI tools and preventing their misuse by unintended parties.

Who is auditing the powerful companies given exclusive access to this AI cyberweapon?

If hobbyists can breach a top AI lab, what hope is there against state-sponsored hackers?

With AI finding decade-old flaws, is any critical software in the world truly secure anymore?

Does this leak prove that keeping powerful AI secret actually makes the world less safe?

Is the AI arms race forcing top labs to quietly abandon their public safety promises?