At Black Hat 2026 in Las Vegas, Orca Security co-founder and CEO Gil Geron discussed how generative AI is dramatically expanding the population of people who build and deploy software—and forcing enterprise security teams to rethink how they protect applications, code, cloud infrastructure, and AI agents.
As developers adopt AI coding tools and nontechnical employees increasingly create applications through platforms such as Lovable, Replit, Vercel, and other AI-assisted environments, Geron argues that security must follow the application wherever it is built.
He explains Orca Security’s expansion into AI application and code security, why context matters more than simply generating additional alerts, and why governance and user permissions will become more important as autonomous AI agents gain the ability to take real-world actions.
Core Takeaways
Everyone Is Becoming a Software Builder: AI-assisted development is expanding software creation beyond professional developers to product managers, analysts, marketers, and other employees, dramatically increasing the number of applications security teams need to understand and protect.
Security Depends on Context: An AI-built application or development platform may be secure on its own, but risk depends on how the application connects to enterprise data, identities, permissions, cloud resources, and other systems.
Orca Advocates a “Best of Platform” Approach: Rather than optimizing for marginally more vulnerability detections, Geron argues that organizations need a unified security platform capable of understanding which vulnerabilities actually matter based on their potential impact across code, cloud, applications, and infrastructure.
AI Agents Require Governance and Permission Controls: As agents become capable of executing increasingly powerful actions, organizations need to constrain them according to user permissions and business context while adopting security tools capable of operating at AI-driven speed.
Key Quotes
AI Is Turning Everyone Into a Builder
“What we’ve seen is that there is a huge expansion in the cohort that produces value for a company. And where natively people that are supporting your business, like developers, engineers that are building your software and deploying it are your main concern, the main concern of the production service of your software. Now the exposure expands to analysts, to product managers, to marketers.”
“And where companies are in such a huge pressure to adopt AI, to get the benefits of AI, it causes companies to try to move as fast as they can with AI. And what we are trying to do is to make sure they’re doing it safely.”
Context Determines Whether an Application Is Dangerous
“The challenge is not about the tool, whether it’s risky or not. It’s about how it is being used. The context is what matters. So let’s say you’ve just deployed a new app in Vercel and Vercel is perfectly secure. But what about if it’s connected to your Salesforce and it exposes all of your Salesforce data to the world?”
“And what if it is allowing you to access protected resources in your environment? So understanding the context of how you use it, how do you deploy, which data is being accessed, what is exposed—it is what matters. And we help you make sure it’s secure.”
Security Teams Need Fewer Alerts—and Better Context
“What our claim is that that 0.5% difference between vendors is not what matters. Context matters. And so understanding that you have an issue that has an impact on your environment that you should fix now is what makes a difference. Companies have tens of thousands of alerts.”
“So the argument that if you just had two more alerts, you would be more secure is ridiculous. Instead of that, let’s focus on the 10 alerts that matter, and you can do it only with context.”
AI Agents Must Operate Within User Permissions
“The issues that start to emerge more and more are around the predictability and around the governance of what’s being deployed. I’ll give you an example. A hotel chain we work with created an agent to do bookings of hotel reservations. Normal. And you ask for it to make a reservation, that’s fine. What if someone asks to make a reservation in all of the hotel rooms in all of the chain?”
“It needs to act in the context of the users, in the permission set of the user. In a very similar way, we are seeing it with AI. We give you a very powerful tool. It’s amazing, but you need to use it correctly.”
