28-05-2025
Establishing A Secure Foundation Of Trust For AI Development
A race car isn't fast because of the engine alone—it's the brakes that make speed safe and ... More controlled. Trust enables acceleration.
Artificial intelligence is changing how we build software. It speeds up development and helps teams ship faster. But with that speed comes a big question: Can we trust the software AI creates?
In a world of AI-powered code, trust isn't a bonus—it's a must.
AI coding tools like GitHub Copilot and Gemini Code Assist are everywhere. Developers are using them to build faster and automate more. But AI also brings new risks.
AI doesn't just help write code. It changes how software is built. It changes who builds it. And it changes what's possible—both good and bad.
I sat down with Danny Allan, CTO of Snyk, to talk about how software development is evolving and what we need to do to ensure we can trust it. 'We're in a perfect storm right now,' he declared.
Allan described the three converging fronts of the perfect storm: AI is creating more code than ever. That code is often less secure than what senior developers would write. And AI-native applications have a larger attack surface, especially when large language models are involved.
A recent study by Snyk found that 96% of CISOs are worried about how AI is being used in development. That concern is well-placed.
AI-generated code may look like regular code—but it's not. The risks are different. That's why we need a new approach.
LLMs add new dangers. Prompt injection, model theft, data leaks and poisoned training sets are all part of the picture. Allan noted we are also still not logging prompt history or tracking model outputs in most organizations.
He compared today's AI rush to the early days of cloud. 'Back then, no one was locking down instances or logging access,' he said. 'Now, we're doing the same with AI models.'
AI isn't just another tool. It's a new layer of infrastructure. And right now, it's going mostly unsecured.
That's where AI trust platforms come in. These tools aim to secure the entire AI pipeline—from how the code is written to how the models behave.
Snyk announced the launch of its own AI Trust Platform to help address this. It includes:
Allan explained the platform's goal: 'Technology can never achieve its full potential unless we trust the technology that we're using.'
The rise of AI coding assistants has sparked fears that software engineers might soon be obsolete. But that vision misses the bigger picture. AI doesn't eliminate the need for developers—it changes what they do and how they add value.
Danny Allan sees a future where developers fall into three evolving categories:
In this model, AI doesn't shrink the developer community—it expands it. Everyone becomes a builder, but with different levels of sophistication and responsibility. And as AI-generated code becomes more common, the need for oversight, security and skilled guidance only grows.
AI is a powerful tool. But human judgment—especially when it comes to security, ethics and edge-case logic—remains irreplaceable. The challenge isn't how to replace developers. It's how to re-skill and redefine them for the AI era.
As AI tools become more connected, through systems like Model Context Protocol, companies must make sure those connections are safe. Snyk, for example, is offering both integrations and security guidance for MCP. That's key. Every new tool is also a new attack surface.
Speed without safety leads to disaster. But trust lets you go faster with confidence.
Allan shared a quote from his CEO to drive the point home: 'The reason why racers can go fast is because they have brakes. It's not because of the engine. You can go faster. And so if you want to trust it, it's the brakes that you're trusting. It's not the engine itself.'
AI will keep changing how we work. That's a good thing. But trust needs to grow with it.
The companies that succeed will be the ones who build trust into every layer—from the models they use to the code they ship. That means educating developers, adopting secure tools and setting clear standards.
AI is the engine. Trust is the brake.
And both are needed if we want to go the distance.
