Anthropic's Mythos model has identified a second vulnerability that is already being exploited in the wild, and according to The Register, it is one that human researchers probably would not have discovered. The finding adds to growing evidence that AI systems can outperform humans in certain security tasks, potentially changing the economics of cybercrime.
The news comes as the FBI announced a third arrest in its investigation of the ShinyHunters hacking group, which breached Bureau systems in late September. The latest suspect is reportedly a Canadian citizen taken into custody in Pennsylvania, following earlier arrests in the Netherlands and Jordan. ShinyHunters has said the attack was not financially motivated but aimed at preserving its reputation, after the FBI made statements about an earlier breach of the Instructure platform.
Together, these developments point to a shifting landscape. Human hacking crews are still being pursued and caught, but AI's ability to discover novel vulnerabilities suggests that future attacks could become more automated and harder to attribute. The Register's discussion noted that the Mythos finding raises questions about how central humans will be to cybercrime going forward.