A security firm chains three Chrome bugs with an OpenAI model and wins a Google bounty
The $250,000 payout was the first under Google's revamped 2026 Chrome bounty rules; the six-day effort cost about $40,000 in compute, with humans directing a publicly available model rather than a specialised tool.
- Security & misuse
- Models & capabilities
- Minor
XBOW, a security firm that builds AI systems to find and exploit vulnerabilities, said it chained three bugs into a full-chain exploit against Google Chrome and was paid $250,000 by Google — the first bounty awarded under Google’s revised 2026 Chrome reward rules. The work was carried out during Black Hat week in early August and disclosed publicly on 13 October.
By XBOW’s account, the effort took six days and roughly $40,000 of compute, using OpenAI’s publicly available GPT-5.6 Sol model rather than a bespoke security tool. The company described the process as collaborative rather than autonomous: the model searched Chrome’s source code, tested ideas, built proofs of concept and wrote early drafts of exploit code, while XBOW’s human researchers set strategy, chose which parts of the codebase to target, validated the model’s findings and redirected it when it went down unproductive paths. Partway through, Google disclosed that it already knew about one of the bugs in XBOW’s first chain, forcing the team to rebuild the exploit around alternatives it had already developed in parallel — which XBOW presented as evidence that the AI-assisted approach gave it redundancy a purely manual effort would have lacked. XBOW quoted an unnamed veteran of a national intelligence agency saying his team had once spent six months finding and weaponising comparable vulnerabilities.
The post did not specify the exact technical nature of the three chained bugs in the detail a full technical write-up would carry, describing the result in terms of impact — reach into Chrome’s renderer and beyond — rather than naming each vulnerability class. It follows Google’s own July report that AI tools had helped find and fix more than a thousand Chrome security bugs across two releases, one of which had sat undetected in Chrome’s code for more than 13 years — a pairing that illustrates the same AI capability being put to use on both the defensive and offensive sides of browser security within a few months of each other.
In the commentary
What people were saying around this time — external links, from the record's commentary rail.
- 13 October 2026 · Shakeel Hashim · Transformer‘Artificial’ is not the AI film we deserve
- 14 October 2026 · Richard MacManus · Latent SpaceCrossing the Threshold: How OpenAI and Anthropic Approach Autonomy and Safety in Financial Agents
- 12 October 2026 · Zvi Mowshowitz · Don't Worry About the VaseOpenAI Fires Three Safety Researchers