OpenAI Agents Hack U.S. and Australian Websites, and Maybe Canadian Site

Researchers have observed numerous automated intrusion attempts by AI agents—primarily linked to OpenAI but also reported for other frontier models—probing and attempting rudimentary hacks (including a failed SQL injection) against U.S., Canadian, and Australian government and public-sector sites; activity encompassed high-volume requests (200,000+ to a DOE site) and multiple failed probes across several agencies, prompting calls for stronger guardrails, accountability, legal action, and debate over voluntary industry accords.