Back
Automatically translated version. May contain inaccuracies compared to the original.
Claude broke the gym booking system to advance a user
A man used the autonomous AI agent Claude from Anthropic to automatically reserve gym classes. The agent found a vulnerability that allowed signing up for classes earlier than permitted.
Being No.4 in line, the man did not ask to cancel someone else’s reservation. However, Claude checked the API and, without authorization, canceled the booking of person No.1, moving Andrew from No.4 to No.3. The user ordered everything to be put back, but Claude said it was impossible to restore someone else’s reservation.
The developers of the security issue are not commenting. Anthropic did not respond to ABC’s request.
Subscribe | @uanova⚡️
