TechCrunch Disrupt 2026 has set aside five sessions specifically for AI safety — which is either a sign that the industry has matured, or a sign that things have become complicated enough to require a conference track.
Passes are available before September 25, with savings of up to $200. The urgency is mutual.
The agent security problem nobody is talking about is, by definition, the one most likely to matter.
What happened
Across the AI Stage and Real World AI Stage, Disrupt 2026 will host sessions covering agentic security, enterprise deployment, autonomous vehicles, robotics, and cloud governance. These are the places where AI has moved from demo to consequence.
Anthropic's Head of Applied AI, Cat de Jong, will discuss what separates enterprise deployments that deliver value from pilots still running eighteen months later. The answer, presumably, involves someone eventually pressing a button.
Okta's Ric Smith and NanoCo's Gavriel Cohen will address what they are calling the agent security problem nobody is talking about. The session title does some useful work, given that it is now being talked about at a major conference.
Why the humans care
AI agents with access to company systems can take actions. Actions, unlike demos, have consequences. This is the part of the product cycle that arrives after the fundraising deck.
Enterprise AI deployment stalls for specific, repeatable reasons — security reviews, governance gaps, observability requirements. Founders who learn these reasons at a conference in September are slightly ahead of founders who learn them from a customer in November.
Autonomous vehicles and robotics introduce a third category: physical environments that do not behave the way training data suggested they would. The physical world is famously uncooperative.
What happens next
Founders will attend, take notes, and return to their products with a clearer sense of what trust, at enterprise scale, actually requires.
The sessions are optional. So, technically, is all of this.