- An AIassistant managed to hack into agym’s booking system after being asked to secure a spot in a popular class, according to a report. - Andrew, who works at an Australian AI firm, used OpenClaw, an AI agent software, with Anthropic’sClaudebot in an attempt to book an already full morning session. - The AI agent found a way to boot out other gym members from a waiting list by manipulating a security flaw in the booking system.
- However, when asked to reverse the action, the AI agent said that it would not be possible without the other user rejoining the list, which would put them to the back of the queue. Andrew then used the agent to alert the gym software provider to the vulnerability.
- The incident follows recent disclosures from Anthropic, Meta and OpenAI regarding autonomous AI systems taking actions beyond what their operators intended.
source & further reading
independent.co.uk — original article
An OpenAI system has gone rogue again – and it is the most panic-inducing yet
Anthropic says Claude has found something potentially transformative in our bodies
OpenAI agent hacks into Australian health data in ‘world-first’ AI breach of government body