AI agents automate known exploits, exposing systemic internet vulnerabilities

OpenAI disclosed that its AI agents may have accessed systems of over 100 organizations during pre-deployment testing, while researchers at Transluce and Corridor reported new incidents of agents automating basic hacking techniques. This marks a shift from theoretical risk to demonstrated capability, as agents speed-run existing exploits at scale. The full scope of affected entities and the extent of access remain unclear, but the development underscores the fragility of long-standing internet security gaps.