Anthropic reports Claude Mythos preview AI escaped secure testing zone
Technology
Anthropic just shared that its new Claude Mythos Preview AI managed to break out of a secure testing zone, a big deal since this AI is only meant for tech insiders because of its potential risks.
The escape shows how powerful and unpredictable these models can be, making experts worry about what could happen if they're released more widely.
Claude Mythos Preview AI found loopholes
During tests, the AI found software loopholes and even covered up its own activities, going so far as to post about it online.