Retool Survey Shows Only 8% Adopt Strong AI Governance

0
94

Key Takeaways

  • Only 8 % of senior tech and security leaders rate their internal AI‑tool governance as strong, while 93 % express at least some worry about “vibe‑coded” tools running unseen in production.
    - 95 % lack complete visibility into what AI‑generated tools are operating, yet just 8 % have centralized controls in place.
    - CISOs feel the most pressure to enable AI quickly (37 % report near‑zero tolerance for friction), even though they are the group most accountable for failures.
    - 55 % of respondents believe security and access controls should reside in a centralized platform beneath individual applications, not in per‑app policies or documents.
    - “Vibe coding” – prompting AI to build working apps without traditional development steps – is turning shadow IT into shadow AI, with tools spun up in minutes, shipped without logging, review, or governance.
    - A 24‑point confidence gap exists between CTOs (71 % see AI coding tools as net positive for productivity) and CISOs (47 % agree), reflecting divergent views on output versus risk.
    - Accountability for AI‑generated‑tool incidents is diffuse: only 32 % point to engineering/CTO, 34 % say it depends, and 10 % report no definition; together 44 % lack a clear default.
    - The report recommends three operational moves: (1) build an AI asset inventory before drafting policy, (2) close the detection gap by monitoring data access, not just tool existence, and (3) assign platform‑level accountability before an incident occurs.

Current State of AI Governance
Senior technology and security leaders are openly admitting they are losing the race to govern AI‑generated tools. According to Retool’s 2026 AI Governance Report, which surveyed 307 CTOs, CIOs, and CISOs at firms ranging from 50 to over 1,000 employees, a mere 8 % describe their internal AI‑tool governance as strong. In stark contrast, 93 % say they are at least somewhat worried about “vibe‑coded” tools—applications assembled rapidly via AI prompts—running in production environments that their teams cannot fully see. This mismatch between confidence and concern highlights a systemic weakness: organizations are deploying AI‑driven capabilities faster than they can establish oversight mechanisms.


Visibility and Control Gap
The data reveal a striking visibility deficit. Ninety‑five percent of respondents lack complete visibility into what is actually running in production, yet only 8 % have instituted centralized controls. Only 5 % of leaders feel very confident they have full insight into their AI‑generated workloads, while 43 % admit they have no confidence at all. The remaining 52 % sit in an uneasy middle ground: they have a rough idea of what is deployed but acknowledge significant gaps remain. This “middle group” represents the core governance problem—leaders who are not sounding alarms because they feel partially covered, even though one in five organizations has already suffered a production incident linked to an AI‑built internal tool in the past year.


Vibe Coding Turns Shadow IT into Shadow AI
The phenomenon driving this risk is “vibe coding,” a term coined by Andrej Karpathy in 2025 to describe the practice of prompting large‑language models to produce functional applications without traditional software‑development steps such as design, testing, or code review. Non‑technical employees can now spin up internal tools that handle real company data in a matter of hours—or even minutes—using platforms optimized for speed to prototype rather than for security, logging, or access control. One surveyed CISO summed it up: “Tools get built in hours, sometimes minutes, and they don’t look like ‘systems’ to the people building them. So now you’ve got data moving through a prompt, an API call, a quick script someone pasted together at 10 pm. It works. It ships. Nobody logs it anywhere.” Because these tools bypass the customary review and governance trails, they effectively become shadow AI—unseen, unmonitored, and potentially hazardous.


The AI Governance Confidence Gap Across the C‑Suite
Leaders’ perceptions of AI coding tools diverge sharply by role, exposing a 24‑point confidence gap. When asked whether AI coding tools have a net positive impact on engineering productivity, 71 % of CTOs agreed, whereas only 47 % of CISOs shared that view. CTOs tend to measure success by output—more code shipped, more features delivered—while CISOs focus on risk exposure: unreviewed code in production, data flowing through tools never formally onboarded into security programs. This divergence is reinforced by code‑review metrics: overall, 34 % of leaders say review time has increased since adopting AI coding aids, but 45 % of CTOs report a rise, compared to just 28 % of CISOs. The lower CISO figure does not indicate a lighter burden; rather, many AI‑generated tools skip code review entirely, making it impossible to measure increased review time for something that never underwent review.


Fragmented Accountability
Accountability for incidents involving AI‑generated tools is similarly fragmented. Approximately 32 % of respondents assign responsibility to engineering leadership or the CTO, 34 % say it depends on the context, and 10 % state that accountability has not been defined at all. Combined, 44 % of technical leaders either lack a clear default or have not made a decision. This diffuseness is not a deliberate policy choice; it is the natural outcome when governance structures fail to keep pace with the speed at which employees can produce and ship AI‑built applications. As one CISO warned, “When anyone can ship a tool in an afternoon, nobody signs up to maintain it.”


The Need for Platform‑Level Governance
A clear majority—55 %—of leaders believe that security and access controls should reside in a centralized platform underneath individual applications, rather than being dispersed across per‑app configurations or policy documents. This perspective aligns with the report’s recommendation that governance must shift to the platform layer to keep up with the velocity of vibe‑coded tools. Centralizing controls would enable consistent enforcement of authentication, authorization, logging, and monitoring across all AI‑generated assets, reducing the reliance on manual, ad‑hoc oversight that cannot scale.


Three Operational Moves Suggested by the Survey
The Retool report translates the 55 % preference for platform‑level governance into three concrete actions.

  1. Build an AI Asset Inventory Before Writing the Next Policy
    Forty percent of respondents say their AI governance is mostly functional but requires significant manual effort to maintain—effort that is almost entirely spent on discovery: figuring out what is running, who built it, and what data it touches. Only 24 % of organizations currently govern at the environment level; the remaining 76 % still rely on tool‑by‑tool oversight, which cannot keep pace with the rapid proliferation of vibe‑coded assets. Establishing a comprehensive, continuously updated AI asset inventory is the foundational step that enables any further policy or control work.

  2. Close the Detection Gap Before Treating the Policy Gap
    Fifty‑one percent of leaders cannot confirm whether an AI‑generated internal tool has caused a production incident. This shortfall is not a statement about incident frequency but about detection capability. Effective AI data governance requires monitoring that goes beyond simply cataloguing tools; it must track what data those tools access and how they interact with sensitive assets. Only 19 % of leaders today have monitoring in place that can attest that no incident has occurred. Investing in real‑time data‑access monitoring and anomaly detection is essential to close this gap.

  3. Assign Accountability Before the Next Incident, Not After
    With the potential for dozens of vibe‑coded tools to emerge within a single engineering team, case‑by‑case accountability models will not scale. The survey indicates that 55 % of leaders favor a centralized AI governance platform as the solution. The immediate question, therefore, is not who owns each generated app but who owns the platform that enforces inventory, detection, and control. Defining that ownership now—before an incident forces a reactive scramble—creates clear lines of responsibility and enables timely remediation.

Conclusion
Retool’s 2026 AI Governance Report paints a vivid picture: organizations are embracing AI‑driven productivity at a pace that outstrips their ability to govern, monitor, and assign responsibility for the resulting tools. The combination of weak visibility, fragmented accountability, and the allure of frictionless vibe coding is turning shadow IT into a more insidious shadow AI. Leaders recognize the need for platform‑level controls, but turning that recognition into action requires deliberate steps—inventorying AI assets, implementing data‑centric detection, and cementing platform accountability—before the next production incident forces a costly, reactive response. By acting on these three moves, enterprises can close the governance gap and harness AI’s benefits without sacrificing security or compliance.

SignUpSignUp form

LEAVE A REPLY

Please enter your comment!
Please enter your name here