Adel El Hallak, Nvidia’s vice president for agentic AI, told Semafor that the scaffolding built around an AI model—the instructions and governance code that determine how an agent operates in practice—is at least as crucial for safety as the model itself.
“Rather than creating fear and FUD, let’s just secure it. We know how. We’ve done these things before,” El Hallak said, referencing the acronym FUD (fear, uncertainty and doubt).
Why the harness matters
El Hallak argued that as AI models become more capable, “harness engineering”—the engineering work that enforces how an agent behaves and what actions it may take—becomes increasingly important. By “harness” he meant the layer of rules and code that governs an agent’s real‑world behavior.
“If you’re training a capable agent, it is important that you don’t put a lion in and expect to be able to control it in a horse’s pen,” he said, emphasizing the need for practical controls.
Industry efforts and regulation
El Hallak declined to state whether government regulation of the agentic layer is required, but pointed to industry efforts such as the Open Secure AI Alliance as an appropriate venue to determine “at which part of the layer do we need to add more strengths and more guardrails and more safeguards.”
His comments align with Nvidia CEO Jensen Huang’s recent position rejecting the idea that the industry must choose between safety and speed, with Huang saying “you could definitely have both at the same time.”
However, some voices in Washington dispute that the private sector can handle these issues alone. Democratic Representative Pete Aguilar said this week, “This is not an industry that can regulate itself,” arguing for federal, industry‑wide AI safety rules.
Why this matters
Agentic AI—systems that can take autonomous actions—are a focal point in debates about AI risk because of their potential to act without human intervention. El Hallak’s technical argument supports the idea that engineering approaches and layered safeguards around agents can mitigate risks, offering a counterpoint to calls for slowing development based on geopolitical or ideological concerns.
“The world’s not going to slow down,” El Hallak noted, arguing that practical safety mechanisms should be developed and implemented as capabilities continue to advance.
