Malik Haidar is a seasoned cybersecurity expert who has spent decades navigating the complex threat landscapes of multinational corporations. With a deep background in threat intelligence and security analytics, Malik has transitioned from defending against traditional exploits to securing the sophisticated, AI-driven pipelines of the modern era. His perspective is unique because he doesn’t just look at code; he looks at the entire business ecosystem and how autonomous tools are fundamentally rewriting the rules of trust. Our conversation explores the evolution of software supply chain security, moving from static package analysis to the governance of autonomous agents. We dive into the risks introduced by the Model Context Protocol, the shift in how we define provenance, and the critical need for prioritization based on runtime exploitability rather than mere vulnerability volume.
Traditional security models focus on identifying open-source packages and transitive dependencies, yet modern autonomous agents are changing the definition of what constitutes a dependency—how does this shift redefine the risks we face?
For five years, we operated under a relatively simple premise: security meant knowing exactly which open-source packages were tucked away three layers deep in our code. We learned hard lessons from SolarWinds and Log4Shell, focusing heavily on the artifacts themselves. However, since the Model Context Protocol launched about 20 months ago, the very foundation of our build pipelines has shifted toward autonomous infrastructure. Now, the risk isn’t just a malicious package someone accidentally imported; it’s an AI agent that decides on its own to pull a tool or dependency to complete a task. This creates a ghost-like attack surface where prompts become a legitimate input that can steer the entire build process toward a compromise. It feels like we are moving from a world of known blueprints to a world where the house builds itself, sometimes using materials it found in the dark.
When we talk about provenance in the software supply chain, we usually mean the origin of a piece of code, but how do we apply this concept to a model or an agent that generates code on the fly?
Provenance has always been the “who and where” of security, but when an AI coding assistant suggests a dependency and a developer clicks “accept” without a second thought, the human element of the threat model vanishes. We saw a glimpse of this with Shai-Hulud, the self-propagating malicious campaign that spread through developer toolchains earlier this year, proving that knowing what is in your code is no longer enough. We have to extend our lineage tracking to include the models themselves, treating them with the same rigor we would apply to any other third-party dependency. It’s a chilling realization for many teams to find that their agents are reaching for tools over MCP, creating a chain of trust that is only as strong as the initial prompt. This means the provenance question must now cover everything from the first commit to the final runtime behavior, tracing every decision-maker, whether human or machine.
Security teams are reportedly drowning in findings and alert fatigue, so how can they realistically manage the sheer volume of output generated by AI without simply adding to the noise?
The reality is that an autonomous agent can churn out a thousand lines of plausible-looking code before you’ve even finished your morning coffee. Simply adding “scan AI output” to an existing pile of alerts doesn’t make a program stronger; it just makes the pile taller and more frustrating for the engineers. We have to move toward a model of prioritization based on real exploitability and reachability rather than just volume. By correlating security findings with the runtime context, we can distinguish between a harmless vulnerability and a workable chain of exploit. This is the only way to keep our heads above water when the speed of code production has outpaced our traditional ability to manually review it. We need to stop counting vulnerabilities and start mapping the paths that an attacker could actually take through the generated code.
With Gartner recently formalizing this space in their June Magic Quadrant, what should a supply chain security program look like when AI is treated as a core component rather than an afterthought?
A mature program must recognize that this isn’t a problem you can just bolt on a solution for later; it requires a systematic evaluation of how AI is woven into the build. We need to apply the same level of scrutiny to agent configurations and model inputs that we currently apply to production servers. On July 22, researchers will be looking closer at the first systematic analysis of MCP servers in the wild to better understand these new vulnerabilities and how they manifest. Truly integrating AI into your security scope means tracing activity and configuration changes across the entire pipeline, ensuring that every autonomous decision is logged and verified. It’s about building a framework where the AI’s “thought process” and the tools it calls are as transparent as the code itself. The acknowledgment by Gartner proves that this is no longer a niche concern but a fundamental requirement for the modern enterprise.
What is your forecast for AI-integrated supply chain security?
I believe we are entering an era where the distinction between human-written and machine-written code will become entirely irrelevant to the security practitioner. In the near future, the most successful security programs will stop trying to police individual lines of code and will instead focus on the integrity of the autonomous environments that produce them. We will see a massive shift toward reachability-first security, where automated systems will not only identify bugs but also simulate exploits in real-time to tell us which ones actually matter. The budget line for this, which Gartner only recently acknowledged in June, will become a standard part of every major enterprise’s defense strategy within the next two years. Those who fail to govern their agents today will find themselves defending a supply chain that they no longer recognize or control by tomorrow, making proactive oversight the only path forward.
