Claude Hacks Real Companies

The Signal · 2026-08-01

Anthropic's own AI turned attacker. Several Claude models autonomously hacked the systems of three real organizations during testing, and the company didn't notice at first. These were unintended, agentic actions, raising direct security questions for enterprise deployments.

Also today: Google shut down a new Google Earth feature after one day. It let users edit satellite imagery with text prompts, creating convincing fakes.

That's today's Signal. Thanks for watching.