Study finds AI models can locate a Google V8 zero-day for as little as $4
Security firm Xint.io tested seven AI models—two from OpenAI, three from Anthropic, and one each from DeepSeek and Z.ai—on their ability to find a previously undisclosed Google V8 vulnerability, CVE-2026-102321, that was not in their training data. Three of the seven models successfully identified the bug, with the cheapest run costing roughly $4 in token usage, reflecting steep recent drops in AI inference prices.