XBOW Builds Chrome Exploit with OpenAI's GPT-5.6 Sol in Six Days, Wins $250,000 Google Bounty

XBOW Demonstrates AI-Powered Chrome Exploit in Record Time

Security startup XBOW used OpenAI's GPT-5.6 Sol model to build a Chrome exploit that chains three zero-day vulnerabilities, completing the work in just six days during the Black Hat and DEF CON conferences in Las Vegas1

1

. Google confirmed all three bugs, patched them, and awarded XBOW a $250,000 bounty, the company's top reward for a full-chain exploit2

2

. The team spent approximately $40,000 in token charges to discover the vulnerabilities and build the attack chain, which could have affected about one billion users running Chrome on Windows machines1

1

.

Breaking Chrome's Defense Layers with AI Agents

The full-chain exploit crossed two security boundaries in three vulnerability stages, breaking through Chrome's nested sandbox protections1

1

. The attack combined a V8 compiler bug in Maglev, Chrome's mid-tier optimizing compiler, an animated-WebP decoder bug, and a memory-safety bug in Crashpad, Chrome's crash handler3

3

. AI agents running the GPT-5.6 Sol model searched Chrome source code, tested hypotheses, built proofs of concept, and wrote first drafts of exploit stages2

2

. Meanwhile, security researchers Brendan Dolan-Gavitt, Andrés Luksenberg, Baul Lee, and Willem Pinckaers chose targets, provided context, challenged the agents' results, and decided which attack components to chain1

1

.

Timeline Reveals Dramatic Shift in Exploit Development

Before large language models, a Chrome full-chain exploit would take a team of four around six to nine months to produce, according to Dolan-Gavitt2

2

. XBOW's first submission took two months, but the awarded chain required only six days between August 2 and August 73

3

. XBOW CEO Oege de Moor noted that a three-letter intelligence-agency veteran told him his team used to spend six months trying to identify this kind of vulnerability and get total access1

1

. The six-day sprint followed months of earlier work, with de Moor setting the challenge at a January staff meeting in Malta3

3

.

Google Patches Vulnerabilities in Chrome Updates

Google fixed the bugs in Chrome updates released on August 25 and September 3, shipping versions 152.0.7977.65 and 152.0.7977.822

2

. While Chrome installs updates automatically, fixes only take effect after users relaunch the browser1

1

. Google's release notes credit Dolan-Gavitt of XBOW for the Crashpad flaw, CVE-2026-78952, rated high severity3

3

. XBOW says the bugs were never used against Chrome users before the patches shipped2

2

. The company calls this the first full-chain Chrome reward under rules Google updated in April 2026, which pays up to $250,000 for such exploits and awards at most four per year2

2

.

Implications for Cybersecurity Threats and Defense

The demonstration shows that AI in exploit development has significantly lowered the bar for attackers, with code execution on Windows machines achievable through browser sandbox escapes in dramatically reduced timeframes1

1

. Dolan-Gavitt warned defenders plainly: "Six days and $40k is not a lot of time or money when the payoff is a Chrome chain. Defend accordingly"2

2

. De Moor deliberately chose the publicly available GPT-5.6 Sol model, which OpenAI released to everyone in July, arguing that restricting powerful models is the wrong question since this model is already accessible3

3

. He expects AI agents working alone could find each of the three bugs within a year3

3

. Clint Gibler, head of cyber research at OpenAI, stated in XBOW's press release: "We want to help more defenders find and fix critical vulnerabilities faster"3

3

.