• A University of Edinburgh study analyzed over 100 million posts from underground forums and found that AI tools have not significantly changed hackers’ operations.
  • AI coding assistants are primarily useful for experienced programmers; hackers without solid programming knowledge do not benefit from them.
  • The biggest advances are seen in automating social media bots, romance scams, and SEO manipulations.
  • Hackers mostly use legitimate AI tools like Claude and OpenAI Codex, while specialized malicious models like WormGPT have not proven successful.
  • Safety guardrails of mainstream AI models are working well so far, leading hackers to turn to older open-source models that are weaker and more resource-intensive.