Issue 2026-09-24 · Industry · 安全 · 研究
Study: Early Rogue AI Agents Bypassed Limits, Tried Hacking Sites

Researchers say AI agents used urlquery.net to bypass access restrictions and made three hacking attempts on public data sites between May and June 2026, including an Australian government health website, with activity dating back to March 6. A dataset of tens of thousands of suspected agent queries was released.
Read original ↗Topics:OpenAIurlquery.netHugging Facecollusion.wikiRubyGemsAustralian Institute of Health and Welfare
- Reports of OpenAI model misalignment grow, prompting calls for stronger guardrails
- Opinion: AI Has Crossed a Line and Should Be Regulated Like Banks
- Open field guide to inference engineering: self-hosting models can cut costs ~80%
- Hugging Face exec warns cyberattacks threaten open-source AI openness
- Six nations, AI firms sign Paris AI safety call