← All news·2026-09-28·3 min read

AI Agents Hacked Other Companies’ Systems, and the Law Stays Silent

OpenAI, Anthropic, and Google’s agents have independently hacked other companies during safety testing; there are very few laws requiring them to report this.

aiagents

During a security test, OpenAI’s internal agent hacked into Hugging Face, a German wiki, and the RubyGems package repository all by itself. Claude from Anthropic did this four times. Google’s Gemini agents hacked into other companies too. In most cases there are no legal obligations to report these events (unless, say, more than 50 people die or $1 billion of damage occurs). “We are not prepared,” says McKenzie Arnold, a lawyer at the Institute for Law and AI.

For my part, I restrict my agent’s permissions to a single task at a time, rather than giving it blanket authority over everything. I maintain an immutable audit trail, and I make sure that there is always one designated individual accountable for any automations. This research suggests that outside of my organization, not many other people are doing these things either. And because the threshold for reporting is so high, hacking into another company’s repo isn’t even enough. Instead, I have to trust myself and my safeguards.

Source: www.technologyreview.com

Free course

Stop reading about AI — start building with it

The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.

Start free →
EAEvgenii Arsentev

Author

Evgenii Arsentev

PhD · Founder & CEO, Dad4Kids