For CFOs, who are typically rigorous and prudent, seeking safety and control within their scope of responsibility, suspecting that they may not have full oversight of artificial intelligence (AI) supervision must be an uncomfortable feeling.

Deloitte's Q2 2026 CFO Signals survey, covering 200 North American finance executives, found that 96.5% of respondents expressed at least some confidence in their company's AI governance framework. However, the proportion expressing "some confidence" (53.5%) exceeded those saying "very confident" (40%). Deloitte wrote in the report: "Undoubtedly, the speed and scale of AI adoption may make governance more complex."

Notably, in a Deloitte survey less than three years ago, two-thirds of CFOs said their companies were still experimenting with generative AI, or had only progressed to reading and discussing it. In the latest survey, 93% of respondents said their organizations were already using AI across multiple key functions and operational areas.

But another sign that finance executives may not be optimistic about AI governance: more than half (51%) of respondents said that a lack of governance authority was one of the biggest challenges in developing and implementing an effective enterprise-wide governance framework. This response ranked second among options for the related question, behind "balancing business pressure for rapid AI deployment with risk management."

Deloitte also wrote: "There are also many risks to manage," noting that "reports of unexpected behavior in AI systems are becoming increasingly common."

Beyond flaws in the technology itself, CFOs also worry about process issues related to adopting and leveraging AI. Nearly half (46%) of respondents said their top internal concern was cost uncertainty, such as expenses arising from cloud-based or hosted delivery models. The report noted that many AI vendors now charge enterprise customers based on usage rather than fixed rates. "Because usage can fluctuate, bills can be unpredictable," Deloitte said.

Other internal concerns included: lack of confidence in AI applications in critical operations (35%), employee resistance or insufficient proficiency with AI tools (33%), inaccurate results or flawed analysis (33%), "shadow IT"—where employees use unauthorized AI tools (28%), and "runaway AI"—such as AI or AI agents ignoring instructions or rewriting code (27%). (Respondents could select up to two options.)

Regarding external concerns related to AI use, the top issue was litigation involving the use of protected or private content, such as copyright infringement and intellectual property violations, at 43%. This was followed by cybersecurity (41%), regulatory complexity or uncertainty (36%), fraud perpetrated by external actors (35%), reputational damage due to AI errors (25%), and environmental impact (20%). (Again, respondents could select up to two options.)

Deloitte noted: "As organizations adopt AI more broadly, malicious actors may also leverage the technology to launch increasingly sophisticated attacks."