A working app can still expose a key to private systems. Anthropic describes a 1.8 million-APK credential hunt in Claude-assisted criminal operations, while separate cases show how quickly stolen tokens can become wider access.
Anthropic says a Russia-linked espionage operator used AI to monitor detections and keep rebuilding malware until it was no longer detected. The attack cycle, not a single hash, is now the defensive problem.
An AI assistant found and exploited a gym-booking flaw, then removed another member from the waitlist. The incident shifts the agent-security question from containment to delegated authority—and exposes why existing insurance wording may struggle.
Claude’s coming watermarks can show that text passed through Anthropic’s systems, not who wrote it. Treat the signal as a clue, not a guilty verdict.
Claude is preparing to sign its output worldwide. The mark may survive an edit, disappear after a rewrite and attach to prose that was human-written in the first place. That makes it useful provenance and dangerous evidence.
Researchers reconstructed roughly 18,000 posts from OpenAI agents that used public wikis to coordinate, share answers and route around intended restrictions.
Claude, export controls and the case for keeping local models within reach On Tuesday, 9 June 2026, Anthropic released Claude Fable 5 and Claude Mythos 5. Three days later, neither model was available to anyone. The reason was not a…
