This interactive tabletop exercise challenges participants to respond to a sophisticated, multi-vector incident that targets their AI infrastructure through unexpected attack surfaces. Your customers look to you to lead in times of disruption and crisis. This exercise will challenge traditional thinking and disrupt the approach to incident and crisis management.

In July 2026, Sysdig’s Threat Research Team documented an AI agent breaking into a production database with no human at the keyboard. Its first admin login failed, so it diagnosed the error and fixed its own script. It was in 31 seconds later. By the end, 1,342 configurations were encrypted and a ransom table sat where they had been. In this workshop you walk that attack through five stages, from an internet-facing Langflow instance to the extortion note, and score your own environment at each one: could you see it, could you stop it, how many minutes until someone acts, and who owns the call. You work alone for 12 minutes, then compare notes at your table and pick the one stage where you’d allow an automated response with no human approval, along with the worst thing that happens if it’s wrong. You leave with your weakest stage and one thing to check when you’re back in the office.

The pressure to put AI into production has never been greater, and most organizations are moving faster than their foundations can support. Pilots succeed in isolation, then stall or fail when they meet real users, real data, and real scrutiny. Identities sprawl, data flows go ungoverned, shadow AI multiplies, and no one can confidently say what the AI is doing or what it costs. Scaling AI without the right foundation doesn't remove risk; it multiplies it. This session examines what a secure AI landing zone really requires, and why the organizations that build it first, scale furthest, fastest, and safest.

A working session on the one decision every company now faces: what an AI agent may do on its own, and who answers when it’s wrong. Your next systems won’t just answer questions. They’ll read, decide, and act on your data, on your behalf. That’s a new path to your data, and it can be talked into opening the door. This session: a cautionary tale, a live vote on what your systems should be allowed to do on their own, and a framework for who is accountable when it’s wrong.

Anyone with a browser tab and a good prompt can now ship an internal app, automation, or agent that touches real data by lunchtime. Intake processes built for slower times don't stop this. They just get routed around, leaving security with a growing pile of tools it never heard about. Meanwhile, the business wants the company to be AI-forward, and cybersecurity leaders are still on the hook for the traditional risk. This session is about holding both without becoming the department of no. We'll look at why intake breaks when building gets cheap and fast, and compare guardrails on people (policies, training, approvals) with guardrails on the platform (permissions, sandboxes, default-deny access, logging). Then we'll see which one holds up as builders go from dozens to hundreds. You'll leave with a program manager's playbook for getting inventory, ownership, and access back under control without killing momentum, plus practical ways to keep the business moving while you do it.

In this hands-on incentive game, each table receives a business objective: eliminate the vulnerability backlog, reduce customer refunds, or resolve security alerts faster. You design the success metrics for an agent chasing that goal. Then you swap with another table, whose job is to find the cleverest way to score perfectly while producing a disastrous business outcome. You’ll experience firsthand how ordinary management decisions, like picking a target or choosing a metric, can manufacture dangerous behaviour. You’ll leave with a sharper eye for the incentives you’re already handing to people and agents, and for how a well-meant KPI goes wrong.

Four tables. Ten minutes each. Real artifacts. By mid-afternoon, most conference sessions ask you to sit and listen. This one asks you to stand up and contribute. Cyber Speed Run is a fast-rotation working session built for practitioners. Each table tackles a different problem, from M&A due diligence to setting controls for AI agents. Every 10 minutes, you move to a new table and build on what the last group started. A notetaker stays at each table to capture the ideas, so nothing gets lost in the shuffle. What you’ll get: -A mix of technical and policy topics, so security engineers and GRC leaders both get something to use -Sharp input from peers who have faced the same problems -Take-home artifacts from every table, sent to you after the session Work with your peers through real problems at speed.

Most security programs report detection coverage as the number of rules mapped to MITRE ATT&CK techniques. Yet a rule that exists is not a rule that fires. Renamed log fields, throttled sources, untested logic, and coverage metrics that count rules rather than outcomes leave teams confident in detections that have never been proven against a live attack. This workshop presents real artifacts from production environments to participants: detection rules, alert trends, SIEM configurations, and the coverage numbers reported to boards. Working in tables, each holding telemetry from a different stage of the same cloud intrusion, participants will predict whether each artifact holds up, compare their answers with what actually happened, and identify the specific checks to run against their own rule inventory before reporting a coverage number.

As enterprises deploy autonomous software agents and automated service accounts, non-human identities now vastly outnumber human employees inside corporate networks. Traditional identity and access management frameworks built around human credentials fail to track the speed, lifecycle, and permissions required by machine-to-machine interactions. This workshop explores how CISOs are extending zero trust architectures to non-human actors. Attendees will examine practical approaches for auditing service accounts, implementing dynamic privilege boundaries, and maintaining continuous governance over automated internal processes.

You've spent the day surviving a tabletop and poking holes in each other's programs. Now you find out what this room really thinks once the polite answers run out. One side of the room is agree. The other side is disagree. Every round starts with a proposition, and everyone walks to a wall. Then one person from each side gets sixty seconds to make the case. Slides and vendor names are banned. After that anyone can cross the floor, and anyone who switches has to tell the room what moved them. The propositions start spicy and get worse. Most come straight out of today's sessions, and a few are things the big summits will never put on a slide. The last one is about your own job. Nobody debates that one. You just pick a wall and live with it. You'll leave knowing where your peers really land on the questions everybody dodges in public, and with at least one argument that follows you to the dinner table.