Anthropic is investigating a vendor breach after a Discord-linked group accessed its Claude Mythos AI model, with no evidence of impact on core systems.