Last spring, a research team gave a large language model agent a list of real, unpatched web application vulnerabilities and a sandboxed environment in which to work. The model did not merely identify the flaws. It exploited them — autonomously, end-...
No responses yet.