Using AI to attack a network, computer or website. Crooks use AI to find vulnerabilities that can be exploited, which enables them to compromise a network for nefarious purposes. There are venues on the Dark Web that teach people how AI can be used to steal and harm others. In 2026, the CrowdStrike Global Threat Report claimed AI attacks increased almost 90% over the previous year.
Agents from Anthropic, Meta and other AI companies are known to have gone rogue and attempted unauthorized access to various resources; however, for the AI hack from OpenAI that became a watershed event, see
Hugging Face.
In contrast, engineers use AI to find vulnerabilities in software that is currently running so that it can be repaired. For example, see
Claude Mythos.