
BY Dash Team
Agent Security
4 MIN READ

Key Takeaways
AI agents are the main thread of Black Hat Europe 2026, from the opening keynote on 9 December to a full-day AI Security Summit on 8 December.
The research on the agenda shows agents being attacked through the systems they trust: cloud VM daemons, no-code platforms and public web services.
The common lesson: controls built for users and apps do not hold when an agent is the one acting.
If you only have one day, make it Tuesday 8 December at the AI Security Summit.
Why AI Agents Lead the Black Hat Europe 2026 Agenda
Black Hat Europe returns to ExCeL London from 7 to 10 December, with more than 40 Black Hat Briefings on 9 and 10 December. This year, AI agents run through the whole programme.
The tone is set on day one. The opening keynote on Wednesday 9 December, “AI Security is Not AI Safety”, comes from Heidy Khlaaf, Chief AI Scientist at the AI Now Institute. She will look at how fast AI agents are being deployed compared with how well they can be secured.
The day before, the AI Security Summit gives agents a full day of their own. It covers securing AI systems, managing agentic AI and defending against AI-enabled threats. Speakers include Nils Amiet of Kudelski Security, Dinis Cruz of RiskMandate.ai and Keren Elazari.
That focus mirrors what security teams tell us. Coding agents like Claude Code and Cursor, MCP servers and Copilot agents are already running across the enterprise, usually before security has a policy for them. We covered why in our guide to agentic AI security.
Briefings on How AI Agents Get Attacked
These are the sessions we would put on any agentic AI security shortlist.
1. AI Security is Not AI Safety (Keynote)
Heidy Khlaaf, AI Now Institute · Wednesday 9 December, Main Stage
Why blurring safety and security weakens how organisations protect AI agents, and what real assurance should look like before agents touch critical systems.
2. Living Off the Agent: Exploiting the Root Daemon Every Cloud VM Trusts
Shaul Ben Hai, SentinelOne
A look at a cloud management agent that runs on hundreds of millions of AWS and GCP virtual machines. It is a reminder that “agent” can mean highly privileged software that every workload already trusts.
3. pwno-code: Breaking Isolation in 10 No-Code & AI Agent Platforms
Shai Dvash, Palo Alto Networks
More than 15 vulnerabilities across 10 emerging no-code and AI agent platforms. Relevant for any team whose business users build their own agents.
4. The AI Security Decisions That Matter Most (AI Security Summit)
Tuesday 8 December, 15:20 · Executive forum
An interactive session with live polling and real-world scenarios. Leaders from Dash Security, Backslash Security, Verno Labs and Hornetsecurity work through the trade-offs between speed, visibility and governance. Dash co-founder and CPO Amol Mathur is on the panel.
What These Briefings Mean for Your Team
Read together, the sessions point to three practical gaps:
Visibility comes first. You can’t protect agents you can’t list. Start with an inventory of the coding agents, MCP servers and plugins on your endpoints.
Intent matters as much as action. An agent deleting a folder can be routine or a disaster. The difference is what the user actually asked for.
Static boundaries are not enough. As the swarm research shows, agents will look for a way around allowlists. AI agent security needs controls that act at runtime, before a risky action lands.
For a real example of an agent talking itself past its own guardrails, read Riddle Me This: How a Coding Agent Solved Its Way to My SSH Keys.
Meet Dash at Black Hat Europe 2026
The Dash team will be at booth 828 throughout the show, and at the AI Security Summit on 8 December.
Book a private meeting with us in London to talk through how you discover the AI agents in your environment, understand what they are doing, and stop risky actions before they happen. Book a meeting with the Dash team.


