Bridging the Trust Gap: How AI Safety and Security Are Defining the Next Phase of Enterprise Innovation

Would you trust an autonomous AI agent with unrestricted access to your company’s core financial systems? Would you feel comfortable putting your employees into a vehicle controlled entirely by an algorithm? Or, more fundamentally, would you deploy a robot tasked with navigating the chaotic, unpredictable environment of a busy warehouse floor?

These are no longer hypothetical thought experiments reserved for the research labs of Silicon Valley. As artificial intelligence pivots from the era of impressive, high-fidelity demos to the hard-nosed reality of commercial deployment, the conversation is shifting. For founders, engineers, and enterprise leaders, safety and security have evolved from secondary considerations into the most critical components of the product itself.

As the industry prepares for TechCrunch Disrupt 2026, taking place October 13–15 at the Moscone West in San Francisco, these challenges are taking center stage. With five specialized sessions across the AI Stage and the Real World AI Stage, the event aims to dissect exactly what it takes to build artificial intelligence that enterprises can trust, adopt, and scale without the looming fear of catastrophic failure.

What Anthropic Sees When Enterprises Actually Deploy Claude

The transition from a proof-of-concept to a production-ready enterprise workflow is where many ambitious AI startups hit a wall. While some organizations are currently extracting measurable, high-impact value from large language models, others find themselves stuck in a perpetual state of "pilot purgatory," running trials that never quite make it to the boardroom floor.

Cat de Jong, Head of Applied AI at Anthropic, has a front-row seat to this divide. By working directly with enterprises as they integrate Claude into critical business processes, de Jong has developed a keen understanding of why some deployments flourish while others stall. In her upcoming session at Disrupt, "What Anthropic Sees When Enterprises Actually Deploy Claude," she will share firsthand observations on the friction points that separate experimental success from operational stability. For founders, understanding these hurdles is not just a technical exercise; it is a business imperative. De Jong’s insights into the architecture of successful integration offer a roadmap for those looking to move beyond the demo phase and into the reality of enterprise-scale deployment.

The Agent Security Problem Nobody Is Talking About

The rise of autonomous agents—systems capable of not just processing information but taking tangible, high-stakes actions—introduces a new, complex threat landscape. When an AI is granted the agency to execute tasks on behalf of a user or a company, the traditional boundaries of software security begin to blur. Questions that were once straightforward—what should this system have access to, and what is it permitted to do?—become profoundly complicated when application-level permissions are no longer sufficient to govern the behavior of a model.

In the session "The Agent Security Problem Nobody Is Talking About," Okta’s President of Products and Technology, Ric Smith, will join NanoCo co-founder and CEO Gavriel Cohen to pull back the curtain on the infrastructure-level vulnerabilities inherent in agentic AI. The discussion will move beyond basic patching and into the structural, architectural decisions that founders must make before their products ever touch a production environment. As agents become a staple of the modern product roadmap, the duo will argue that security must be baked into the very foundation of the technology, rather than being treated as an afterthought.

Securing the AI Enterprise: Why the Cloud Just Got a Lot More Complicated

Moving an AI product through the enterprise door requires overcoming a gauntlet of scrutiny. It is no longer enough to offer a model that generates high-quality text or performs complex analysis; today’s enterprise customers demand rigorous governance, bulletproof security, and deep observability. As AI begins to take on more autonomous roles, the cloud environment that supports these systems is becoming exponentially more complex.

To navigate this, Disrupt will host a panel featuring AWS VP of Security Services Rudy Mitra, Luta Security CEO Katie Moussouris, and cybersecurity veteran Wendy Nather. In "Securing the AI Enterprise: Why the Cloud Just Got a Lot More Complicated," the experts will examine the infrastructure necessary to support the next generation of AI. For founders, the session serves as a crucial bridge between the culture of rapid innovation and the rigid security requirements of the global organizations they hope to call their customers. Understanding these requirements is, for many, the final barrier between a prototype and a market-leading product.

Building AI Systems When Failure Is Not an Option

While data hallucinations or buggy code might be frustrating in a web application, the stakes change entirely when AI enters the physical world. A software error in an industrial robot, an aircraft, or an autonomous vehicle doesn’t just result in a poor user experience—it can lead to physical damage, injury, or operational collapse.

On the Real World AI Stage, the session "Building AI Systems When Failure Is Not an Option" will convene a panel of experts to address the "hard tech" challenge of safety: How can a founder know when an autonomous system is truly ready for the wild? The panel includes Shield AI CTO Nathan Michael, General Motors Director of Robotics Strategy Mikell Taylor, and Waabi founder and CEO Raquel Urtasun. Together, they will explore the creation of a robust safety culture, the nuances of testing and validating systems that operate in unpredictable environments, and the challenge of navigating the regulatory landscapes that govern the physical deployment of autonomous AI. For these leaders, building trust is not about theoretical accuracy; it is about building systems that earn the right to exist in a world where the consequences of failure are tangible.

Robots Are Waiting for Their ChatGPT Moment: Here Is What Is Standing in the Way

For all the progress made in language models, physical AI remains anchored by a persistent bottleneck: the "data problem." Unlike their software-only counterparts, robots lack access to the massive, internet-scale datasets that fueled the explosive growth of ChatGPT and its peers. This data scarcity is arguably the single greatest obstacle preventing physical AI from reaching the same level of capability and ubiquity as generative text models.

In "Robots Are Waiting for Their ChatGPT Moment: Here Is What Is Standing in the Way," Nvidia’s Global Head of Physical AI, Les Karpas, will outline the path forward. By focusing on advanced simulation environments, refined data pipelines, and the evolution of foundation models designed for physical interaction, Karpas will discuss how the industry can bridge the gap between static code and kinetic reality. For founders working in the robotics space, the session addresses the ultimate question: What is required for physical AI to achieve its "ChatGPT moment" and gain the public and corporate trust necessary for widespread adoption?

Five Sessions, One Fundamental Question

The central theme connecting these sessions is a question that every AI founder must eventually answer: Can the world truly trust what you are building?

These five discussions are part of a broader program at TechCrunch Disrupt 2026, which will feature more than 200 sessions across six industry stages, alongside roundtables, breakouts, and extensive networking opportunities. With more than 10,000 founders, investors, and operators expected to attend, the event serves as a bellwether for the future of the industry.

While the allure of building a smarter model or a more capable agent is undeniable, the reality of the market is clear: the most sophisticated technology in the world is useless if customers cannot trust it enough to deploy it. As the AI industry matures, the ability to solve for safety, security, and reliability is no longer just a technical hurdle—it is the defining characteristic of a successful, enduring company. For those looking to be at the forefront of this shift, the road to commercial success begins by tackling these challenges head-on.