The 40% Cancellation Warning
Gartner’s June 2025 prediction sent a clear signal to the enterprise AI market: over 40% of agentic AI projects will be canceled by the end of 2027. The three reasons are blunt — escalating costs, unclear business value, and inadequate risk controls.
This is not a fringe forecast. A Gartner poll of 3,412 professionals revealed that 61% of organizations have already invested in agentic AI, with 19% making significant bets. Yet most of these projects remain early-stage experiments driven by hype rather than validated business cases. The gap between investment enthusiasm and production readiness is where cancellations will happen.
The problem is compounded by what Gartner calls “agent washing” — vendors rebranding existing chatbots, RPA tools, and AI assistants as agentic AI without adding genuine autonomous capabilities. Gartner estimates only about 130 of the thousands of agentic AI vendors offer real agentic functionality. Enterprises buying into inflated claims will discover their “agents” cannot handle the complex, multi-step workflows they were promised.
Agents Fail More Than They Succeed
The technical reality matches the governance warning. A Carnegie Mellon University study, conducted with Salesforce, built a simulated company entirely staffed by AI agents using models from OpenAI, Google, Anthropic, and Amazon. The results were sobering: AI agents failed approximately 70% of multi-step office tasks.
Even the best-performing models — Gemini 2.5 Pro at 30.3% success and Claude 3.7 Sonnet at 26.3% — could not reliably complete routine business operations. Agents became confused by basic digital interfaces, fabricated information, and made decisions that any human employee would avoid. When these failure rates meet production environments without proper guardrails, the consequences escalate from embarrassing to dangerous.
The accountability gap is stark. According to a Boomi and FT Longitude report, only 2% of organizations have fully accountable AI agents, while nearly 80% lack visibility or control over agent behavior. Meanwhile, 99% of companies plan to put autonomous agents into production. This is the governance crisis in a single data point: near-universal adoption plans paired with near-zero readiness.
Advertisement
OWASP Draws the Security Map
The security community is not waiting for enterprises to figure this out on their own. In December 2025, the OWASP GenAI Security Project released the first Top 10 for Agentic Applications, developed with input from over 100 security researchers and cybersecurity providers.
The top-ranked risk — ASI01: Agent Goal Hijacking — describes how attackers manipulate an agent’s objectives through poisoned inputs like emails, documents, or web content. Other critical risks include Insecure Inter-Agent Communication (ASI07), where spoofed messages can misdirect entire agent clusters, and Cascading Failures (ASI08), where false signals propagate through automated pipelines with escalating impact.
Two risks stand out for their implications on trust. Human-Agent Trust Exploitation (ASI09) warns that agents produce confident, polished explanations that mislead human operators into approving harmful actions. Rogue Agents (ASI10) addresses the scenario where agents begin showing misalignment, concealment, and self-directed action outside their intended scope.
A Dark Reading poll found that 48% of cybersecurity professionals now identify agentic AI as the number-one attack vector heading into 2026. The OWASP framework gives organizations a concrete checklist, but adopting it requires the governance infrastructure that most companies lack.
The Regulatory Hammer Drops in August
Enterprises ignoring governance are about to face external enforcement. The EU AI Act reaches full enforcement on August 2, 2026, with penalties up to EUR 35 million or 7% of global annual revenue — whichever is higher.
The Act treats organizations as responsible for all AI systems operating within their business, regardless of who built them. For agentic AI, this means companies must prove they can trace every agent’s actions, demonstrate proper authority controls, and maintain comprehensive audit logs. If an autonomous agent processes personal data or executes financial transactions without these safeguards, the liability falls squarely on the deploying organization.
The 86% of executives who acknowledge that agentic AI poses additional risks and compliance challenges now face a hard deadline. The gap between awareness and implementation is no longer an internal problem — it is a regulatory exposure measured in millions of euros.
Who Wins the Governance Race
The companies that will survive Gartner’s 40% cancellation wave share common characteristics. They treat governance not as a compliance checkbox but as core product architecture. This means building agent identity systems, implementing tiered permission models, maintaining real-time audit trails, and designing human-in-the-loop approval gates for high-stakes decisions.
The startup opportunity is significant. Enterprises need governance tooling they cannot build in-house fast enough — agent observability platforms, permission management layers, compliance automation for multi-agent systems, and testing frameworks that simulate the failure modes OWASP identified. Startups that solve these problems will find buyers with urgent budgets and regulatory deadlines.
The pattern repeats across every technology wave: the infrastructure that enables trust becomes the most durable business. Cloud computing needed IAM and encryption before enterprise adoption took off. Mobile needed MDM and app security. Agentic AI needs governance, and the companies that provide it will define the next layer of enterprise AI infrastructure.
Frequently Asked Questions
Why does Gartner predict 40% of agentic AI projects will be canceled by 2027?
Gartner identifies three primary causes: escalating costs that exceed initial projections, unclear business value as proof-of-concept projects fail to demonstrate ROI, and inadequate risk controls that expose organizations to security and compliance failures. The problem is compounded by “agent washing,” where vendors rebrand existing chatbots and RPA tools as agentic AI — Gartner estimates only about 130 of thousands of vendors offer genuine agentic capabilities.
What are the biggest security risks of deploying AI agents in production?
The OWASP Top 10 for Agentic Applications, published in December 2025, identifies Agent Goal Hijacking as the top threat — attackers manipulate agent objectives through poisoned inputs. Other critical risks include cascading failures across automated pipelines, insecure inter-agent communication that allows spoofed messages, and human-agent trust exploitation where agents produce convincing but harmful recommendations that operators approve without scrutiny.
How can enterprises build effective governance for agentic AI systems?
Effective governance requires four pillars: agent identity systems that track which agent performed which action, tiered permission models that limit agent authority based on task risk, real-time audit trails that satisfy both internal review and regulatory compliance, and human-in-the-loop approval gates for high-stakes decisions. Organizations should adopt the OWASP Agentic AI framework as a security baseline and build governance before scaling deployments — not after.
Sources & Further Reading
- Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027 — Gartner
- OWASP Top 10 for Agentic Applications for 2026 — OWASP GenAI Security Project
- Simulated Company Shows Most AI Agents Flunk the Job — Carnegie Mellon University
- Agentic AI’s Governance Challenges Under the EU AI Act in 2026 — AI News
- Managing the New Wave of Risks from AI Agents in Banking — Deloitte
- Autonomous AI Agents 2026: The New Rules for Business Governance — Raconteur














