Autonomous AI has crossed the threshold from experiment to enterprise priority. The models are accurate enough. The business case is clear. Boards are pushing for speed. But for most CIOs, security concerns — not technical limitations — are holding them back.
Indeed, 77% of more than 11,000 CIOs in Gartner’s 1H26 CIO Report said security and risk were the biggest barriers to scaling autonomous technologies. Their concerns are legitimate: Autonomous AI agents can leak data, make costly mistakes and create audit nightmares.
But a growing number of CIOs are finding ways to move fast without sacrificing security. The answer, they say, lies in guardrails, governance and a different kind of partnership among CIOs and security and privacy executives.
Rising pressure from autonomous AI
Approximately eight in 10 executives expect autonomous business to dominate their industry by 2030, according to an April 2026 Gartner survey of 469 CEOs. Boards and leadership, recognizing the competitive stakes, are pushing CIOs to keep up with the shift.
“My executive team and board are relatively fluent in AI,” said Keith Fulton, chief data officer at Jack Henry, a software provider for financial institutions. “They see the potential. They’re saying, ‘How can we go faster?’ I’m saying, ‘I want to go faster, too. But we want to be careful.'”
That tension between business pressure and security discipline is playing out across industries. A Sembi survey of nearly 3,800 software development leaders found that 86% say security issues delay releases at least occasionally, and 63% cite privacy and security concerns as barriers to AI adoption.
“There’s real pressure from every part of the business to accelerate AI adoption right now, and security [and IT] teams can’t respond to that pressure by simply saying no,” said Rinki Sethi, CISO at Upwind Security and former security leader at Twitter, Rubrik, and BILL. “The conversation has shifted from blocking innovation to enabling it responsibly.”
Top issues with agentic AI
Data exposure. The core issue, multiple leaders say, is visibility — or the lack of it. “Most organizations still don’t fully understand what AI agents have access to, what actions they’re capable of taking or how they behave once deployed into production environments,” Sethi said. “Data exposure is a major issue, particularly when agents can access internal systems or move information across environments without clear controls.”
At Jack Henry, Fulton draws a distinction between cybersecurity and data security. “It’s not really cybersecurity” that is the biggest concern, he said. “It’s the security of money and data. We need to make sure PII doesn’t leave the building when we’re talking to hyperscale agents.”
Agent fallibility. The challenge is compounded by the fallibility of agentic AI. “The agents have gotten to 80-99% accuracy. They’re getting better, but they’re not 100%,” Fulton said. “If you had an Excel spreadsheet and 1% of what it returned was a made-up number, no one would use it. That’s where agents are today.”
Shadow AI. Adding to the complexity is the rise of shadow AI. “Employees are adopting AI tools because they improve productivity, and most security teams are discovering usage after the fact rather than through formal approval processes,” Sethi said. “The answer isn’t banning everything because that usually drives activity further underground.”
Security in AI from the start
The organizations moving fastest aren’t bolting security on after deployment. They’re building it from the beginning.
“[Security] has not been a brake for us,” said Chase Christensen, segment CIO and vice president of enterprise solutions at Jabil, a global manufacturing services company. “It only slows things down when we don’t design security into our processes. We really make sure we design security upfront within our [software development lifecycle] SDLC. That removes all the hurdles and allows us to scale quickly.”
Christensen has also reframed how Jabil thinks and talks about shadow IT — and AI. “We don’t talk about shadow IT — we talk about democratization of IT,” he said. “Enterprise IT can be slow. Our job is to enable platforms. When we provide the right data and rules around consumption, what looks like shadow IT becomes a growth engine for the organization.”
Sethi agreed that early integration is essential. “The organizations doing this well are treating AI systems like production workloads from day one, rather than experimental side projects,” she said. “Retrofitting security after deployment rarely works because by that point the AI system is already integrated into workflows, APIs and data environments that are difficult to untangle.”
Set the right AI guardrails: The dog park principle
Jack Henry’s Fulton has embraced guardrails not as constraints but as accelerants.
“I come back a lot to the analogy of a dog park,” he said. “I take my puppy to the dog park because I want her to have freedom, but I don’t want her to run in the street. She sees the fence and doesn’t go beyond it. She can be playful, and I know she’s OK. The key to going fast is having the right guardrails.”
Risk level determines guardrails. Those constraints are risk-calibrated at Jack Henry. “We have a rubric for judging the risk level of actions an agent might take,” Fulton said. “Depending on the risk level, we apply different guardrails. Money movement is very hard to undo. The guardrails for that have to be very careful and rigorous compared to those for a copilot used to help write a Word doc.”
Accountability is nonnegotiable. “Every agent has to be tracked, audited and traced to a single human being responsible for its behavior,” Fulton said, driven in part by federal regulations. “You can’t send an agent to jail. Every action of an agent has to be traced back to a person responsible for it.”
Continuous visibility into AI deployments
For Sethi, the biggest shift is moving from static policy reviews to runtime monitoring.
“Security becomes a brake when teams rely on traditional governance models that weren’t built for real-time, autonomous systems,” she said. “The organizations moving fastest are the ones building visibility and runtime context into AI deployments from the beginning instead of trying to bolt controls on later.”
That means redefining what “good enough” security looks like. “If you can’t answer what data an agent can access, what actions it can take, or whether its behavior has deviated from normal patterns, you’re not ready to scale,” Sethi said. “The mistake is treating AI deployment as a one-time security review rather than an ongoing monitoring commitment.”
The changing CIO-CISO relationship
Multiple leaders point to the CIO-CISO dynamic as a critical enabler — or a bottleneck — when it comes to autonomous AI.
“AI has made the CIO-CISO relationship much more operationally intertwined,” Sethi said. “Historically, security and IT could operate on parallel tracks, but autonomous systems force much tighter coordination because infrastructure, data governance, application development and security are now deeply connected.”
The conversations have changed, as well. “It’s less about compliance checklists and more about operational resilience, visibility and managing business risk at the speed of automation,” she said. “In many organizations, the CIO and CISO are now jointly responsible for enabling AI safely, rather than treating security as a downstream approval function.”
Chief privacy officer’s role. At Jack Henry, Fulton said the traditional CIO-CISO partnership is only part of the picture. “The CPO may be more involved than the CISO,” he said. “It’s about respecting the privacy of clients and customers — and not trusting hyperscalers with that data.”
The organizations scaling autonomous AI aren’t ignoring security. They’ve just stopped letting it be the reason that nothing ships.
“Don’t wait for perfect governance before moving forward, because the business will outpace you,” Sethi said “Speed without visibility creates risk, but visibility gives you the confidence to move faster responsibly.”
Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.