Are ChatGPT, Claude or Gemini Useful in a Breach Investigation? with Josh Lemon

Threat actors are already using large language models to accelerate the early stages of a compromise. Josh Lemon wanted to know what the same tools could do for the responders on the other side.Recorded at AdelaideSEC 2026, David Savva-Willett speaks with Josh Lemon, Chief of Digital Forensics and Incident Response at SoteriaSec and a SANS principal instructor, immediately after his talk on that question. Josh has been putting the major models in front of forensic evidence for years, and reports real movement: where they once sent an inexperienced analyst off in entirely the wrong direction, they now answer factual questions reliably. What they still lack is the creativity to ask why something is wrong, or what a threat actor is likely to do next.The discussion ranges across where LLMs are genuinely saving hours in a SOC, from one off Python tooling to threat intelligence summaries to executive status updates; a response Josh received that read unmistakably like vendor advertising; why he warns his students that an AI written report may one day be read by an expert witness in court; how MCP servers are being folded into SANS forensics classes; and whether any of this helps with the on call burden and burnout that have dogged the profession for years.It closes on the exchange every responder knows by heart, where no evidence of exfiltration gets heard as no exfiltration, and Josh's observation about what a model eager to please would say if a lawyer asked it the same question.