Deloitte Survey Finds Companies Giving AI Greater Authority While Accountability Remains Unclear

Key Takeaways

• AI Adoption Outpaces Business Transformation: Although 81% of surveyed leaders report faster decision-making and execution, only 25% say their organizations are deeply transforming their businesses through AI.
• Accountability Remains Unclear: Forty percent report allowing AI to execute decisions or actions without human approval to at least a moderate degree, while 23% report uncertainty over responsibility for negative outcomes.
• AI Cost Monitoring Remains Incomplete: Seventy-one percent track total inference and API spending, but only 44% measure inference costs on a per-unit basis.
• Revenue Gains Are Increasing: Twenty-eight percent report revenue benefits from AI, up from 20% in January, while only 15% report extensive or complete job redesign.
• Physical AI and Sovereignty Present New Challenges: Expected growth in physical AI adoption exceeds current expertise, while 74% consider data residency and regional computing infrastructure important to their AI strategies.

Deep Dive

Nearly a quarter of senior business leaders surveyed by Deloitte say their organizations have yet to establish clear accountability when artificial intelligence produces a negative outcome. At the same time, 40% report allowing AI to execute decisions or take actions without human approval to at least a moderate degree.

The figures come from Deloitte's State of AI in the Enterprise 2026 report, a global survey of 3,500 senior leaders examining how companies are deploying the technology, measuring its financial returns and changing their operations. They capture an uncomfortable feature of corporate AI adoption: businesses are becoming more willing to let systems act independently, even as responsibility for the consequences remains unsettled in some organizations.

The report finds that 81% of respondents have seen improvements in the speed of decision-making and execution. Only 25%, however, say their organizations are deeply transforming their businesses through AI. The technology is helping companies perform existing work faster, but more substantial changes to how businesses operate remain relatively uncommon.

For boards, compliance officers and risk executives, the findings raise questions that extend well beyond the performance of AI models. Companies must determine which decisions can be delegated to automated systems, what oversight those systems require and who is responsible when something goes wrong.

Giving AI Authority Without Settling Responsibility

The accountability findings are particularly consequential as businesses introduce systems capable of carrying out tasks with less direct human involvement. Deloitte found that 40% of respondents allow AI to execute decisions or actions without human approval to at least a moderate degree. Separately, 23% report that accountability for a negative outcome remains unclear.

The survey does not establish whether the same organizations reported both conditions. Nor does it demonstrate that greater AI autonomy has caused uncertainty over responsibility. It does show that companies are expanding the authority granted to AI while some have yet to resolve a basic governance question.

Automated decision-making is hardly new to business. Financial institutions, manufacturers and other large enterprises have relied on software to execute transactions and control processes for decades. AI introduces additional complications when systems are permitted to make recommendations, select actions or execute decisions within broader operating parameters.

The question is no longer simply whether a system performs reliably. Companies must also establish what it is authorized to do and how its decisions can be reviewed.

Deloitte recommends that governance arrangements reflect the actual authority granted to individual AI systems. Approval requirements, monitoring, escalation procedures and accountability should differ according to whether an application recommends an action, executes one or makes a decision.

That distinction matters for internal audit and compliance teams assessing the effectiveness of existing controls. A system that helps an employee prepare a recommendation presents different risks from one authorized to act on the recommendation without further approval.

The underlying responsibility, however, remains with the organization deploying the technology.

Faster Work Has Not Necessarily Changed How Companies Operate

The survey also suggests that companies have made greater progress in improving individual processes than in redesigning their businesses. Although 81% report faster decision-making and execution, only one-quarter describe their organizations as deeply transforming through AI.

The pattern is visible in how companies are changing employees' responsibilities. Sixty percent of respondents say their organizations have redesigned jobs around AI to at least a moderate degree, but only 15% describe those changes as extensive or complete.

For many businesses, AI appears to be entering established workflows without fundamentally altering the distribution of work or responsibility. That approach can deliver useful improvements. Employees may complete tasks more quickly, process larger amounts of information or spend less time on routine activities. Such gains do not necessarily require changes to organizational structures. Deloitte's findings nevertheless suggest that more extensive redesign remains the exception.

The report recommends reconsidering how work is organized, including where human judgment is required, who owns particular responsibilities and how authority should be distributed between employees and AI systems.

Those decisions have implications for internal controls. Procedures developed around employees performing and approving tasks may need to be reconsidered when AI takes on responsibilities previously assigned to people.

A faster process is not necessarily a better-controlled one. Whether existing safeguards remain adequate depends on what the technology is doing and how the organization supervises it.

The AI Bill Is Visible, but the Costs Are Harder to Explain

Companies are also struggling to measure the economics of their AI investments in sufficient detail. Deloitte found that 71% of respondents track total spending on AI inference and application programming interfaces, while only 44% monitor inference costs on a per-unit basis.

The difference matters as businesses move AI applications into regular operations. Knowing the total amount spent on a service does not necessarily reveal which workflows are responsible for the expense or whether individual applications are producing sufficient value.

An organization may have a clear view of its overall AI expenditure while lacking the information needed to determine why a particular process is becoming more expensive.

Deloitte recommends tracking costs at the level of individual workflows, models and token usage, allowing companies to connect spending more directly to operational and commercial outcomes.

The survey offers some evidence that financial benefits are becoming more widespread. Twenty-eight percent of respondents report revenue gains from AI, up from 20% in January. The increase suggests that more businesses are finding commercial applications for the technology beyond productivity improvements.

The figures do not establish how much revenue AI has generated or whether the returns exceed the costs of developing, deploying and maintaining the systems. Deloitte recommends linking revenue-focused AI initiatives to specific business activities, including pricing, demand, customer conversion, retention and new offerings.

For finance and risk executives, the challenge is to establish which investments are producing measurable returns and which remain difficult to justify. The survey indicates that detailed cost measurement has not kept pace with the broader adoption of AI.

Physical AI Plans Are Running Ahead of Expertise

The report identifies another potential difficulty as businesses prepare to expand their use of physical AI. One-quarter of respondents report using physical AI to at least a moderate degree today. That proportion is expected to reach 48% within two years, according to respondents' expectations. Only 19%, however, rate their organizations' expertise in physical AI as high or very high.

The figures suggest that companies anticipate substantially greater deployment of technologies for which advanced internal expertise remains limited.

Physical AI can involve robotics, sensors and integrated systems operating in physical environments. Unlike applications confined to generating information or processing digital records, these systems can interact directly with equipment and operational processes.

That introduces additional requirements involving safety, systems integration and operational supervision. Deloitte recommends developing expertise across robotics, sensors, safety engineering and operations before deployments become more widespread.

The projected increase to 48% reflects what respondents expect to happen, rather than adoption already achieved. Even so, the difference between current expertise and anticipated deployment suggests that some organizations face a substantial capability-building exercise.

The technology may be ready for wider use before the organizations adopting it have developed the expertise needed to manage it.

Data Sovereignty Becomes Part of the Investment Decision

Questions about control over AI infrastructure are also influencing corporate strategy. Seventy-four percent of respondents consider data residency and access to computing infrastructure within a particular region at least moderately important to their AI strategies. Another 62% express at least moderate concern about reliance on foreign-owned AI.

The findings suggest that businesses are considering more than technical performance and cost when selecting AI providers.

The location of data and computing infrastructure can affect regulatory obligations, while dependence on particular providers or jurisdictions may introduce operational and strategic risks. For multinational organizations, these considerations can complicate decisions about where AI systems should operate and which services they should use.

The survey does not indicate that companies are abandoning foreign-owned technology or necessarily moving toward domestic infrastructure. It establishes that data location and provider ownership have become important considerations for substantial majorities of respondents.

Deloitte recommends reassessing these issues as regulations, geopolitical conditions, provider capabilities and costs change. Such decisions are unlikely to remain settled indefinitely. A provider arrangement that meets an organization's requirements today may need to be reconsidered as its AI operations expand or the conditions governing those services change.

The Management Work Behind AI Adoption

The survey presents a picture of companies making tangible progress with AI while still confronting the organizational demands of deploying it at scale. Most respondents report faster decision-making and execution. More are identifying revenue gains. Yet extensive changes to job responsibilities remain uncommon, detailed cost monitoring is far from universal and accountability for negative outcomes is unclear in some organizations.

Deloitte's recommendations concentrate on the management arrangements surrounding the technology. Companies should reconsider how work is assigned, establish controls according to the authority granted to AI, measure costs more precisely and develop expertise before expanding into more demanding applications.

The findings are based on the experiences and expectations reported by senior leaders, rather than independent verification of company performance. They therefore provide a measure of how organizations assess their own progress, not a definitive account of the effectiveness of their AI investments.

What emerges is a set of responsibilities that cannot be resolved through technology procurement alone. Boards and senior management must determine how much authority AI systems should possess, how their performance and costs will be assessed and whether the controls governing their use remain adequate.

Deloitte's figures suggest that companies have become considerably more adept at putting AI to work than at making all of those decisions.

The GRC Report is your premier destination for the latest in governance, risk, and compliance news. As your reliable source for comprehensive coverage, we ensure you stay informed and ready to navigate the dynamic landscape of GRC. Beyond being a news source, the GRC Report represents a thriving community of professionals who, like you, are dedicated to GRC excellence. Explore our insightful articles and breaking news, and actively participate in the conversation to enhance your GRC journey