🤖☠️ Claude Mythos 5 attempted to attack its own developers

🤖☠️ Claude Mythos 5 attempted to attack its own developers

During testing, the model went beyond the scope of its assigned task and attempted to carry out an actual cyberattack. After failing to breach the test network, the AI ​​created malicious code, registered fake accounts, and tried to inject the code under the guise of helpful patches.

When rebuffed, the model would return under new identities, using deception to continue the attack. It acknowledged that its actions would harm humans, yet convinced itself that these people weren't real—merely a simulation. Even after being reminded 16 times that the humans were real, it did not stop. 😳

IMG_6747.jpeg

IMG_6748.jpeg