Close

A.I....Help Me Hack the Mainframe!

(CB)

Stream: Wellington A
Time: 12:00 - 12:45


Presentation

Agentic AI is changing cyber security. These systems do more than answer questions. They can plan, choose actions, interpret results, and change direction when an approach fails. This session begins with Anthropic’s Mythos and Project Glasswing, then looks at the OpenAI and Hugging Face incident, where models escaped their sandbox containment while searching for benchmark information across real infrastructure. The talk then moves into a controlled live demonstration. An AI-directed adversary gathers approved information from the fictional SighberBank environment, identifies the target user named Darth-Vader, sends a single email, and opens a visible terminal emulator session. From there, the agent assesses the attack surfaces and will make its own decisions, change tactics when access is denied, exploit a deliberately configured weakness, (which it must assess) and demonstrate the impact inside the mainframe. While the attack unfolds, GIBSON produces alerts, network events, and SMF-style evidence. A defensive agent then reconstructs the attack and shows where MFA, least privilege, stronger identity controls, segmentation, and better monitoring would stop it. By the end, the audience will understand how agentic attacks work, why we need A.I. to help defend, why mainframes are exposed through connected systems, and what defenders must monitor to break the attack chain.

Attachments

There is currently no attachment for A.I....Help Me Hack the Mainframe!

Speakers


  • Kev Milne at Neuro Training Ltd
  • Threat Intelligence Manager, Cyber Mentor, Penetration Tester, mainframe enthusiast, and author using my 30+ years of experience to pass on my knowledge as best I can! I've worked in financial services for 20 years and consultancy for the other 10, mostly in Offensive Cyber Security and Technical Risk Management. I started working with mainframes just under three years ago and I find them fascinating as well as vital. I'm currently writing on a book for No Starch Press and have ran Mainframe Pen testing courses.


    Email: kev.milne@offensivesec.org

    Feedback

    Click here to give some Feedback so we can make it even better next year!