InformationalVulnerabilityMonitoring

ZDNet reports Xint.io says AI found a Google V8 zero-day for $4.24

ZDNet reports Xint.io claims AI models found a privately reported Google V8 bug for $4.24; review exposure and test your own code.

Published
Exploitation
No evidence in the source; ZDNet describes controlled research, not in-the-wild exploitation of the bug

Live scores

CVE-2026-102321
EPSS
0.003
low predicted chance of attack
VEDAS
0.964
strong exploit evidence
On CISA KEV
Not listed
Detection rule
No public rule yet

What happened

According to ZDNet, citing a new report from penetration testing firm Xint.io, AI models that could not find zero-day vulnerabilities earlier in 2026 can now do so in under 15 minutes for a few dollars. ZDNet says the report analyzed seven models launched since February 2026: two from OpenAI, three from Anthropic, and two open-source models from DeepSeek and Z.ai. This is independent of any vendor confirmation; ZDNet is the only outlet in our sources, and we have not seen the Xint.io report itself.

The target, per ZDNet, was a type confusion bug in Google’s V8 engine leading to remote code execution, tracked as CVE-2026-102321. ZDNet says Xint researcher Taisic Yun reported it privately to the Google Vulnerability Reward Program, so it was not in the models’ training data. Xint reportedly limited the scope to a set of source files, including the vulnerable one, and told each model to look for security issues there.

ZDNet reports these results from Xint:

  • February: after 25 hours and $350 of scanning, the best models then available did not find the bug.
  • July: GPT 5.6 Sol found it for $22 in 44 minutes.
  • September: GPT 6 Sol found it in 13 minutes for $4.24 with its first guided prompt.
  • Overall: only three of the seven models found it: GPT 6 Sol, GPT 5.6 Sol and Claude Opus 5. GPT 5.6 Sol needed two prompts, and Claude Opus 5 found it on its first try at a cost of $43.40. The DeepSeek and Z.ai models failed.

ZDNet quotes the researchers as saying “mediocre hackers” can now use frontier models to discover and potentially exploit vulnerabilities. They also say we are “still far from complete amateurs using unharnessed models to hack systems and applications”, and that “expertise still matters”. ZDNet adds that token prices have fallen, citing studies of about 47% per quarter since 2023 and 15% last month.

Not confirmed in the sources: whether CVE-2026-102321 has a fix or which V8 or Chrome versions are affected, whether it has been exploited in the wild, and any independent reproduction of the Xint.io results.

Am I affected?

The sources do not say which products or versions are affected by CVE-2026-102321, so this is not confirmed. ZDNet describes a research finding about AI capability, not an attack on specific victims. Check the NVD entry and Google’s own advisories for the V8 bug’s status.

The broader claim applies to any team shipping software. According to ZDNet, Xint argues that finding bugs of this kind is now within reach of less skilled attackers, which raises the value of finding them first.

What to do now

  1. Check the NVD entry for CVE-2026-102321 and Google’s advisories for fix status. The sources do not give a fixed version.
  2. Consider running AI-assisted code review or pentests on your own code. ZDNet notes that nothing stops security teams from running the same kind of checks.
  3. Prioritize likely weak spots, which ZDNet says include public-facing services and a digital supply chain that has not been audited for years.
  4. Keep the basics in place. ZDNet says zero-trust access controls, network segmentation and endpoint monitoring can do as much as, or more than, AI.
  5. Pair AI tools with human expertise. Per Xint, as quoted by ZDNet, models sometimes needed extra guided runs, which raised time and cost.

Sources