The Verge · AIMay 12, 2026, 00:09 UTC+8

Google Stopped a Zero-Day Hack It Says Was Developed With AI

Google stopped a zero-day hack that it says was developed with AI

Google says GTIG found and stopped its first AI-developed zero-day exploit, and the attackers planned a mass exploitation event to bypass two-factor authentication on an unnamed open-source web-based system administration tool; the post does not disclose the tool name.

Why it matters: This hits HKR-H/K/R: Google-backed AI zero-day claim, a concrete 2FA bypass target, and clear security resonance. Missing tool name and exploit details keep it in the 78–84 band.

Read the original ↗Export Markdown

Take

Google’s story reads more like a security narrative upgrade than a sudden attacker capability jump. GTIG’s concrete evidence is a Python exploit with a “hallucinated CVSS score” and textbook-style formatting. The target was a 2FA bypass in an unnamed open-source web-based admin tool, planned for mass exploitation. The tool name, vulnerability details, model, and generation path are not disclosed.

I buy “LLM-assisted exploit development.” I don’t buy “AI autonomously discovered a zero-day” from this snippet. Code models have already been good at PoC adaptation, glue code, and debugging loops for a year; Claude and GPT-class systems lower the scripting bar fast. But zero-day work still depends on target knowledge, a test harness, and validation cycles. The practical signal for defenders is uglier: more mediocre exploit code at scale, with AI fingerprints that are weak evidence by themselves.

Source: The Verge · AI · theverge.com
Published: May 12, 2026, 00:09 UTC+8 · May 12, 2026