What 200 CISOs and CTOs Reveal
At A Glance
- Network and firewall changes for AI deployments take an average of 8 days from request to implementation, and 54% of organizations report delays of a week or more.
- Risk and compliance approvals and cross-team dependencies are the top contributors to those delays.
- Underneath the slowdown is a confidence problem: 85% of security leaders aren’t fully confident their current tools can protect AI deployments in the first place.
- Shadow AI is compounding the caution: 90% of leaders are concerned about unsanctioned AI use they can’t fully see or govern.
Every executive sponsoring an AI initiative eventually asks the same question: why is this taking so long? The model was ready weeks ago, the use case was approved…and yet, the project is still sitting in a security review queue.
We recently surveyed 200 senior security and technology leaders, CISOs, CTOs, CAOs, and CIOs at companies with 1,000 or more employees in multiple industries on how they’re navigating the security risks of AI deployment. The data shows that deployment delays aren’t rare. Rather, they’re the predictable result of a confidence gap plaguing AI security today.
Why Are AI Deployments Taking Longer Than Expected?
The data confirms what a lot of AI executive sponsors are already living through. More than half of organizations report that network and firewall changes alone add one to two weeks to AI deployments, with an average of 8 days lost from request to implementation for a single change. The top contributors, tied for the lead, are risk and compliance approvals (security reviews, change advisory boards) and cross-team dependencies across networking, security, cloud, and application teams.
That delay compounds fast. Every new model, agent, or data source an AI initiative touches can trigger another round of the same change management cycle, and every handoff between teams adds another point where the request can stall. For an executive sponsor, this is the part of the AI security story that rarely makes it into a board deck. It’s not just about breach risk; it’s about how many weeks a vital AI initiative sits waiting on approval before it ever reaches production.
Why Don’t Security Leaders Trust Their Own AI Defenses?
The deeper issue is what’s driving all that caution in the first place. 85% of respondents are not fully confident that their current security solutions can adequately protect their AI deployments. Only 15% report being very confident, and that gap isn’t distributed evenly across roles either. The leaders closest to actual security risk report even less confidence than their technical counterparts, which tracks: it’s the security team’s job to stay skeptical, while the people building and scaling the technology tend to focus more on shipping than scrutinizing its defenses.
That distrust has a direct line to the delays. When a security leader doesn’t fully trust the tooling behind an AI deployment, every change request gets more scrutiny. The confidence gap isn’t separate from the deployment bottleneck; it’s the reason the bottleneck exists.
How Is Shadow AI Adding to the Deployment Bottleneck?
Concern about shadow AI, the unsanctioned use of AI tools outside formal oversight, is close to universal, with 90% of leaders describing themselves as moderately or very concerned. That’s not surprising once you consider human behavior: when an approved AI tool doesn’t fully meet what employees need, they find another way to get the job done.
But from a security leader’s chair, that visibility gap means every new, sanctioned AI request gets evaluated against the assumption that there’s already unmonitored AI activity somewhere in the environment. That assumption drives more conservative reviews, which drives longer approval cycles, which lands right back on the executive sponsor’s timeline.
How Much Pressure Are Organizations Under to Secure AI Right Now?
None of this is happening in a vacuum of low stakes. The vast majority of security leaders, 78%, report high or very high organizational pressure to secure their AI deployments, and this pressure holds up across virtually every industry surveyed, though the intensity varies by sector.
For an executive sponsor, the pressure to move fast on AI has never been higher, and the confidence to do it safely has never felt thinner. Something has to give, and right now, it’s deployment speed.
How to Close the AI Deployment Confidence Gap
Read collectively, this isn’t a story about security teams being slow or difficult. It’s a structural problem: the tools built to secure human access over the last decade weren’t designed for the volume and pace of machine-to-machine change that AI deployments require, so every request routes through a review process built for a different era.
Closing that gap isn’t just an investment in security, but in velocity. The organizations that solve the non-human identity problem won’t just reduce breach risk; they’ll ship AI initiatives faster because security stops being the bottleneck. That’s the gap NetFoundry’s Identity-First Reachability™ is built to close, giving every AI agent, API, and machine workload its own verifiable identity so security teams can approve changes with confidence instead of caution, without adding new firewall rules or VPN exceptions to the queue.
If you’re building the business case for your next AI security investment, or trying to explain to your board why deployment timelines keep slipping, the full data set, including how these numbers break down by industry, company size, and role, can help you build your case.
Download the Full 2026 State of Secure AI Connectivity Report
Frequently Asked Questions
Why are AI deployments taking longer than expected?
Network and firewall changes required for AI deployments typically route through formal change management processes involving multiple teams. According to NetFoundry’s 2026 survey of 200 CISOs and CTOs, these changes add an average of 8 days from request to implementation, with 54% of organizations reporting delays of a week or more, driven primarily by risk/compliance approvals and cross-team dependencies.
What is shadow AI?
Shadow AI refers to AI tools or applications that employees adopt on their own, without formal approval or oversight from IT or security. We see this compound existing bottlenecks: 90% of security leaders are moderately or very concerned about shadow AI, largely because it operates outside the visibility needed to monitor and govern it.
Why don’t security leaders trust their AI security tools?
Many security leaders feel that existing tools, built for traditional human-user access, weren’t designed for the way AI agents connect and operate. At NetFoundry, we see this most clearly in the gap between how confident organizations are securing human users versus machine workloads, which our survey found is nearly a 10-to-1 difference.
What does Zero Trust mean for AI security?
Zero Trust is a security model built on the principle of least privilege: nothing is trusted by default, and every connection must be explicitly verified. We apply this same principle to AI agents, APIs, and machine workloads through Identity-First Reachability™, giving every non-human identity its own verifiable, governable identity rather than relying on shared credentials.
