A.I. All the Things

Marcus Carey opens by describing himself as a former AI skeptic who now uses large language models every day, and frames the talk around answering common objections to AI in security. He argues that the non-deterministic output of LLMs frustrates traditional data science teams, but that for security practitioners the greater risk is avoiding the technology altogether.

In the middle section he points to areas where he sees value: identifying and quantifying risk, detection and response, crisis communications, governance documentation, and tabletop exercises. He stresses keeping a human in the loop and suggests security staff learn basic Python, data preparation and how models are trained, pointing to public datasets and libraries as starting points. He warns that rushed AI adoption and vibe-coded software without security review create new exposure, and calls non-human identities an open problem.

For the closing demos, he shows a browser assistant gathering research on local security leaders, and an MCP server that connects a coding assistant to threat-intelligence and scanning APIs such as Shodan and VirusTotal. He also recounts using a coding assistant to build a ransomware test for a customer, noting that the assistant refused until the request was reframed. The takeaway is to adopt these tools with domain knowledge and keep people in control of the results.