With agentic AI technology advancing at an exponential rate and Fortune 500 firms giving entire global workforces access to autonomous virtual assistants, one question keeps surfacing: Who is responsible when something goes wrong?
The concerns reached fever pitch after Anthropic CEO Dario Amodei's We Must Pace the Frontier letter warned of a commercial race to the bottom on AI safety. Enterprises that have shifted from GenAI to agentic AI face their own version of the problem, and the accountability question is spreading from the C-suite to the front office.
The stakes are already measurable. Researchers find that AI agents fail at roughly 7 in 10 professional tasks, and legal exposure for enterprises through internal and external partners is growing. While CIOs tend to take the lion's share of the blame for failed initiatives, other departments and teams will face increasing exposure, according to Lynn Comp, head of global sales and go-to-market and vice president for the AI Center of Excellence at Intel.
Comp explores these issues as a technology executive and hands-on practitioner, and offers a practical and sobering perspective.
“Don't build agentic AI technology for a demonstration,” she said, quoting a recent phrase that stuck with her. “Build for the deposition instead.”
Agentic AI accountability is the practice of assigning clear human ownership for the decisions and actions of autonomous AI agents, with responsibility scaling to match the blast radius of the outcome.
How Should Accountability Scale?
Comp believes agent responsibility should ultimately mirror how organizations have always handled consequential decisions. The scale and impact of the decision should determine the seniority of the person who owns it.
“I think of the process as no different than core decision making,” she said. “You have decisions that interns are allowed to make. The more senior the decision-maker, the more impactful the decision. That's never changed.”
She grounds the analogy in semiconductor development, where the stakes are viscerally clear. A CEO owns the call to build a billion-dollar fab. An intern owns the Verilog code for a simple function. The same logic applies to AI agents: a market analysis agent that scopes a competitive summary carries a low blast radius, while a financial services agent that makes loan recommendations carries enormous regulatory and legal exposure.
“When you're making decisions on who can get credit and why, you need provenance behind that judgment if you get challenged later,” Comp said. “As the consequences get bigger, that's where the decision makers get more senior.”
Who Owns an Agent When Its Builder Leaves?
Agent identity and ownership remain unsolved problems, Comp said. Most popular agent frameworks tie agents directly to individual users' credentials, creating a structural vulnerability most enterprises have not confronted.
“If you build a brilliant agent that reports on the health of your entire global fleet, what happens when you leave the company?” she asked.
Agents built under an employee's user ID and embedded in operational workflows become corporate infrastructure over time, but the governance frameworks to treat them as such do not yet exist.
“There has to be disambiguation between the author of the agent, their user ID permissions and the fact that it's now corporate property,” Comp said. “I don't think we've grappled with those questions yet.”
She explained that the next stage will give agents their own identities, permissions and lifespans, with a human commissioning and decommissioning them.
“But we're really not even in the second stage of letting chaos reign yet,” Comp said.
How Do Enterprises Rein In Shadow AI?
The current state of enterprise AI agent governance is chaotic, Comp said, paraphrasing a maxim from former Intel CEO and “Only the Paranoid Survive” author Andy Grove: “Let chaos reign, and then rein in the chaos as things get clearer.”
Most organizations are still in stage one, she said, watching what their lines of business build before asserting control. The pattern mirrors cloud adoption: first, teams put cloud instances on corporate cards. Then IT noticed, secured better deals and centrally managed the most useful services.
“I suspect what will happen with agentic AI is very much like what happened with cloud,” she said.
The dark side is shadow AI, in which employees use unauthorized tools within corporate environments. When a free open-source agent gained sudden traction, one enterprise manager reacted with weary recognition: “Oh, what fresh hell is going to hit my fleet?”
Detecting such unauthorized use is a growing concern.
“How are you detecting it?” Comp asked. “What are the markers? What's the telemetry?”
She is optimistic nonetheless, recalling years when IT directors walked office floors with Wi-Fi detectors hunting for rogue access points. The technology changes. The need for oversight does not.
How Can Governance Keep Up With Agent Swarms?
For routine, bounded tasks, human oversight is achievable and the accountability chain is clear. The problem grows qualitatively harder when agents operate in swarms, networks of agents making real-time decisions across volumes of data no individual human can process.
“There's no way a human can have all the context necessary in every decision an autonomous system makes,” Comp said. Swarms widen the gap further. Governance itself must become partially automated to account for the shift.
“Autonomy and governance has to be code,” she said.
The technical layer will not replace human accountability. Courts have been unambiguous that assigning responsibility to the AI itself fails. Instead, a software framework must surface the right information at the right moment for the human who makes the final regulatory call.
Why Does Owning Your AI Force Better Decisions?
Beneath the accountability questions is a more fundamental one Comp finds energizing: What does it mean for an organization to own its own artificial intelligence?
Building market intelligence agents crystallized something she had not expected.
“It forced me to write down how I made decisions, which made me a better manager because I wasn't giving my team enough of a decision framework,” she said.
At the enterprise level, the same dynamic plays out across the organization.
“Owning your own intelligence is forcing us to write down the process by which we come to answers,” Comp said, “and challenge ourselves on the basis of that judgment.”
Making judgment explicit and encoding it in systems that can be audited is the foundation of responsible AI agent deployment. Organizations that can articulate their decision frameworks can also defend them. In the emerging regulatory environment, the ability to stand behind a decision is fast becoming the difference between a defensible outcome and a deposition gone wrong.
Related:
Scott Steinberg is a business strategist, award-winning professional speaker, trend expert and futurist. He's the bestselling author of Think Like a Futurist; Make Change Work for You: 10 Ways to Future-Proof Yourself, Fearlessly Innovate, and Succeed Despite Uncertainty; and Fast >> Forward: How to Turbo-Charge Business, Sales, and Career Growth. Find him at www.FuturistsSpeakers.com and LinkedIn.
© 2026 Nutanix, Inc. All rights reserved. For additional information and important legal disclaimers, please go here.
Related Articles
