OpenAI and Anthropic are amplifying vulnerability discovery capabilities while simultaneously launching cybersecurity defense products.

November 6 — according to a post by blocmates, data from Google's Threat Intelligence Group (GTIG) shows that the number of disclosed vulnerabilities this year has risen from 5,045 in January to 10,477 in July, nearly doubling; the number of exploits has also nearly doubled, rising from an average of about 10.5 per month in 2025 to about 18 per month this year.

The post states that OpenAI recently used an autonomous AI pipeline to produce 948 working Linux kernel root exploits in a single run, consuming about $1.2 million in API credits (roughly $1,279 each), and reproduced 1,724 vulnerabilities in about 30 hours. At the same time, OpenAI launched the cybersecurity program Daybreak for developers, and Anthropic also has the Glasswing program (launched in summer with Claude Mythos); both currently offer subsidized credits and access. blocmates commented that the two companies, while building models that can mass-hunt vulnerabilities and write exploits, are also selling defense solutions — "the arsonist has started selling smoke alarms."