
Organizations are struggling to align who holds power with who bears responsibility in artificial intelligence governance. Authority over key decisions is consolidating at the executive level, while the day-to-day work of governing AI is spreading across the enterprise. This split creates practical risks. When staff bear responsibility but lack authority, they cannot act. When executives hold authority without responsibility, they face no full accountability.
New titles are not fixing the gap
Despite the persistence of issues, specific roles are on the rise. A 2026 study by IBM’s CEO found that 76% of organizations surveyed now have a chief AI officer, up from 26% in 2025. Other titles, such as chief AI agent officer and chief regulatory officer, are emerging, indicating C-suite expansion. However, 79% of executives surveyed are decentralizing decision-making, distributing accountability as AI’s role grows. Having a title does not necessarily mean owning the responsibility.
The confusion deepens with overlapping titles. The role of chief governance officer is already in use, but its meaning varies widely. At 84% of companies surveyed by the Society for Corporate Governance, the CGO function resides within the legal department or general counsel’s office. This differs from South Korea’s SK Telecom, which uses the title within its AI governance structure. These different definitions can introduce more confusion than clarity.
CIOs face a more pressing question than just assigning titles. They need to know who can make distributed controls function together. Written AI policies establish expectations, but they cannot independently stop an agent from accessing restricted information or executing an unauthorized transaction. This disconnect means that consequential approval often happens when someone grants credentials rather than when a governance committee authorizes deployment.
AI agents complicate this arrangement because they can operate across applications, access sensitive information, and initiate transactions without repeated human intervention. To keep up, organizations are moving AI governance from policy documents into the operational systems that enforce decisions. “In most enterprises, AI governance lives in two places that rarely talk to each other: a policy PDF owned by a committee, and a set of OAuth scopes, service accounts, and API keys configured by whoever shipped the integration,” said Bapuji Koraganti, a cybersecurity engineer at a large American mass media, telecommunications, entertainment and tech company.
The shift in handling permissions is key. An agent allowed to access financial systems should not be permitted to change supplier banking information or release payments without restrictions. Patrick Parker, chief innovation officer at Omada, emphasizes that restrictions must be enforced when actions occur, not just described in policy. He adds that an enterprise can have a full library of AI policies yet lack the connection between governance decisions and enforcement.
Insurance and contracts raise the stakes
Procurement and legal departments face pressure to enact governance strategies through vendor contracts. Data-use provisions, audit rights, and liability allocation are sensitive issues requiring renewed attention to protect enterprises from undue risk. Savio Lobo, CIO at Ensono, argues that AI vendor agreements should be part of enterprise risk management, with procurement, legal, security, and technology departments collaborating to determine acceptable risk.
Contractual restrictions alone cannot prevent unauthorized AI actions. Enterprises must connect these obligations to technical controls and operational oversight. Lobo emphasizes that AI vendor review should not be separate from third-party risk management, requiring additional AI-specific questions that integrate with enterprise controls.
