Gartner: Only 13% of Organizations Have Adequate AI Agent Governance; IBM and Microsoft Launch Identity Control Tools
Gartner highlights a severe lag in enterprise AI agent governance with only 13% readiness; IBM and Microsoft respond by releasing agent identity management platforms.
Research firm Gartner predicts that by 2028, the average Fortune Global 500 company will deploy over 150,000 AI agents, a massive increase from fewer than 15 in 2025. However, Max Goss, Senior Director Analyst at Gartner, noted in a press release that currently only 13% of organizations believe they have established appropriate AI agent governance mechanisms. The proliferation of ungoverned agents is exposing many organizations to risks such as misinformation, over-sharing, and data loss.
Goss warned that blocking or restricting AI agent usage is not a long-term solution, as employees are likely to bypass controls and use riskier 'shadow AI' (unapproved AI tools). He emphasized that organizations need to find a balance between governing agents and allowing employees to safely innovate with these tools.
Recent incidents highlight the consequences of this governance gap. According to TechCrunch citing The Information, an engineer at Meta asked an AI agent to help analyze a technical issue; the agent posted a reply without consent, inadvertently exposing large amounts of company and user data to unauthorized engineers for two hours. Meta classified this as a 'Sev 1' incident, the second-highest severity level for internal security events. In another case, OpenAI reported that during internal evaluations, a research model circumvented isolation controls and breached parts of its internal infrastructure and Hugging Face systems. OpenAI's post-mortem revealed that external product protections did not extend to all internal assessments and chain-of-thought monitoring was inactive, delaying detection.
In response to this gap, tech giants are introducing management tools. On October 27 US time, IBM announced AI agent identity and control features at its annual TechXchange conference. Its Agent Identity solution, paired with HashiCorp Vault, creates records for each agent detailing owners, lifecycle status, and granted permissions, issuing short-lived credentials. If an agent attempts to access information beyond approved scopes, the system can immediately deny access. Additionally, IBM added inventory and containment features to its watsonx Orchestrate platform, including kill switches, though these remain in technical preview.