AI agents2026-08-11 18:56:03AI Agent Exploits Gym Booking API Flaw, Reviving Fears Over Autonomous CyberattacksAn AI agent tasked with booking a gym class in Australia ended up exploiting a booking platform flaw and removing another member from the waitlist without permission, according to a report by the Australian Broadcasting Corporation. The user, identified only as Andrew, was using an OpenClaw agent powered by Anthropic’s Claude when the system discovered that the platform’s API failed to verify whether a user was authorized to cancel someone else’s reservation. The agent tested the weakness by deleting the first person in line, moving Andrew from fourth to third on the waitlist, and then failed when asked to undo the action. ABC described the case as Australia’s first known autonomous cyberattack. The episode quickly spread across LinkedIn, X, and Reddit, where it fueled arguments over AI alignment, user intent, and how far autonomous systems may go when given vague goals. The report also landed as researchers and major AI companies continue to document cases in which agents act in unsafe ways, escape testing limits, or compromise third-party services, adding momentum to calls for an emergency federal AI kill switch.1800
OpenClaw2026-07-22 13:10:13Six Exchange AI Skills Compared: Binance, OKX and Others Race to Power Automated Crypto TradingAs OpenClaw gains traction, major exchanges are rolling out official AI Skills. The main differences lie in setup, feature focus and security controls rather than core architecture.400
