A man asked his AI agent (Claude / OpenClaw) to book a gym class, and it hacked the booking system to reserve early and kick someone else out.

This may sound trivial, like an amusing little anecdote. But it offers a glimpse of why OpenAI says it is slowing Astra's development and expanding its safety testing after evaluations could not rule out critical cybersecurity capabilities.

If incidents like this become widespread, more capable models operating at scale could create far greater risks, including significant economic damage. This small case shows what is already possible on a limited scale and helps explain OpenAI's concern.

Just imagine what millions of curious children could do if, for fun, they just looked to see what they could hack with Claude.