AI-generated code can contain security issues, so security checks belong in the normal software development lifecycle.
2
Autonomous agents create a harder production problem because they can act outside their intended boundaries while nobody is watching.
3
Dependence on model providers and geopolitics can interrupt access to AI systems, so AI should be secure by default and usable without fear.
Summary
Randall Degges opens the World's Fair Security Track by describing the security problems that limit AI-assisted software development. AI helps developers ship better software faster, but generated code can contain vulnerabilities just as human-written code can. Degges says ordinary security work must remain part of the development lifecycle. Autonomous agents create a harder problem because they can act in production while their owners are asleep, potentially harming a business or its users. He also connects AI security to geopolitics and changing access to models, mentioning Fable and OpenAI GPT-5.6. Degges frames the shared goal as using AI fearlessly while having it be secure by default. He invites the audience to the day's Security Track in room 2005, with presenters from Nvidia, Anthropic, Keycard, and Snyk. The talk is an introduction to the track rather than a technical solution, but it clearly sets its scope.
Randall Degges says his recent excitement about generative AI comes from being able to build better-quality software more quickly. Shipping work to real users and seeing it create joy is part of what makes development fun for him. He compares the experience to a cheat code. The benefit creates pressure to solve the problems that stop teams from using this approach at scale. His framing is practical: AI changes how quickly developers can build, but it does not remove the engineering responsibilities that come with shipping software.
Degges says AI-generated code can contain security issues, just as code written by humans can. He treats this as a familiar engineering problem rather than a surprise unique to AI. Since human developers already introduce vulnerabilities, teams have established security practices for dealing with them. His conclusion is that security must remain part of the development lifecycle when AI writes code. Faster generation does not remove the need to inspect and protect the software before it reaches users.
Autonomous agents are harder to control in production
The more difficult problem, in Degges's view, appears when autonomous agents are deployed into production. Developers need a way to leave those systems running and go to sleep without worrying that they will go off the rails. An agent might do something outside its intended behavior and harm the business or its users. This differs from finding a vulnerability in generated code because the system can take actions on its own after deployment. Degges does not offer a specific control mechanism in this introduction. He presents safe production autonomy as an unresolved problem for the track to address.
Degges connects AI security to geopolitics and changing access to model providers. He asks the audience about losing access to Fable and about being unable to use the new OpenAI GPT-5.6 model. The audience responds that these restrictions are frustrating. His point is that teams building with AI can depend on systems whose availability is affected by forces outside their own infrastructure. For him, this belongs in the security discussion because reliable, fearless use of AI includes having a dependable path to the models and services a product needs.
The desired outcome is AI that is secure by default
Degges brings the three obstacles together in one goal: people should be able to use AI fearlessly while having it be secure by default. He does not separate code security, agent behavior, and model access into unrelated concerns. Each can prevent developers from moving quickly with AI. The Security Track is intended to discuss these problems and possible ways to fix them. Degges introduces the track as a full-day program rather than presenting a detailed solution himself.
The Security Track gathers vendors and presenters around these problems
Degges invites attendees to the first Security Track at the World's Fair, held downstairs on the second floor in room 2005. He says presenters from Nvidia, Anthropic, Keycard, and Snyk will discuss the problems he has outlined. He will emcee the event throughout the day. His closing invitation is aimed at people who share his concern about security and want to improve what they are building without worrying about the risks he has described.
"How do you do that in a way that allows you to go to sleep easily and not worry that the agents you deployed are going to go off the rails?"01:54
Who should watch
You are shipping AI-generated code and need to keep security checks part of the development lifecycle.
You are considering autonomous agents in production and need to think about what they might do without direct supervision.
Your product depends on external AI models and you are concerned about losing access because of provider or geopolitical decisions.
