AI
Five AI Safety Sessions TechCrunch Disrupt 2026: Building Systems People Will Actually Trust
As AI moves from labs into production, founders face mounting challenges around security, trust, and reliability—five key conference sessions tackle the issues separating demos from deployable systems.
TechCrunch — AI
Real AI deployment brings real consequences. Whether it's entrusting an agent with access to your systems, putting employees in an autonomous vehicle, or deploying a robot that must navigate unpredictable physical environments, the stakes of AI moving beyond experimentation are now forcing founders and enterprises to reckon with safety as a core product requirement.
TechCrunch Disrupt 2026 dedicates multiple sessions across its AI and Real World AI stages to these challenges. Five standout panels offer founders a front-row look at what separates safe, deployable AI from systems that remain stuck in pilot mode or fail in the field.
Enterprise Deployments: What Separates Pilots from Production#
One of the starkest divides in AI adoption is between companies running pilots indefinitely and those delivering measurable business value. Anthropic's Head of Applied AI Cat de Jong works directly with enterprises deploying Claude into mission-critical workflows and sees firsthand where implementations succeed and where they languish.
In her AI Stage session, "What Anthropic Sees When Enterprises Actually Deploy Claude," de Jong will share lessons from the front lines of real-world deployment—revealing the patterns that distinguish projects that gain traction from those that stall. For founders pitching AI solutions to enterprise buyers, understanding these dynamics becomes essential to closing deals.
Autonomous Agents and Infrastructure Security#
AI agents capable of taking action multiply security risks exponentially. The questions compound: What systems should an agent access? What actions should it be permitted to take? And crucially, are application-level permission models sufficient, or is infrastructure-level security required?
Okta's Ric Smith and NanoCo's Gavriel Cohen will jointly tackle "The Agent Security Problem Nobody Is Talking About" on the AI Stage. Their discussion dives into architectural vulnerabilities and the design decisions founders must make when deploying agentic systems. For any company with agents on its roadmap, embedding security thinking from the start becomes non-negotiable.
