Proof-of-Concept in 15 Minutes? AI Turbocharges Exploitation

An AI-powered offensive research system developed by two Israeli cybersecurity researchers has produced more than a dozen exploits, shrinking the time to create proofs of concept to under 15 minutes in many cases. The pipeline uses prompts to a large language model and CVE advisories, with Anthropic's Claude-sonnet-4.0 handling analysis, test applications, and exploit validation against both vulnerable and patched versions of target software.

The project, dubbed Auto Exploit, generated exploits for 14 open‑source vulnerabilities, and the researchers say hands-on tweaking is still required, though the method can deliver results quickly. They note that the low cost—about $1 per exploit—could enable broad exploitation of more vulnerabilities, potentially increasing the number of N-day exploits.

Defender strategies may need to move to machine-speed responses, since the 192‑day median time-to-exploitation in 2024 cited by VulnCheck could shrink as AI-augmented exploitation becomes more common, according to the researchers. According to VulnCheck, nearly 40,000 vulnerabilities were reported in 2024, with only 768 exploited, highlighting the pressure on defenders if AI accelerates exploitation. 29 August 2025.