How to Adopt AI Agents Securely at Scale

When Vercel disclosed its recent breach, the entry point was an AI agent.It’s the new playbook for attacks: An agent gets static and over-scoped access - in this case to Vercel’s Google Workspace. Once the agent was compromised, the attacker walked straight in.

Marco van Hurne call this: “Russian roulette with AI agents.”, And he's been warning about exactly this pattern for years.

Marco has 20 years of experience in Machine Learning. He worked in the field before many of us couldn’t even spell the word “agentic”. Today, he runs the “AI factory” at ASML (world's leading supplier of lithography systems for the semiconductor industry) and lectures on process automation in the Netherlands.He’s recently ran field research with 177 companies that are running autonomous AI agents. His outcome: 70% of high-value agent projects still fail - amongst others due to security concerns. And he keeps coming back to the same fix: stop treating AI agents and AI tools like software. Onboard them like new employees.

Join Marco and Johannes Keienburg, Founder and CEO at Cakewalk, for a working session on what it actually takes to adopt AI agents securely at scale, before the next breach disclosure has your name on it.

What you'll learn

• Why Vercel-style cascades (employee installs AI tool, tool becomes attack vector, customer data walks) keep happening, and what governance would have caught it
• The four zones of AI automation, and why most agent projects fail above zone 2
• The agents-as-employees framework: scoped credentials, proper onboarding, revocation when the job ends
• Why even a "low-risk" zone 1 agent reading your email can damage trust at machine speed, if you don’t get agent governance right
• How to build the observability layer auditors are starting to ask for under DORA, ISO 27001, and SOC 2
• Why human-in-the-loop approval breaks down after the tenth prompt in a row

Even if you can't attend live, sign up anyway and we'll send you the recording.