AI Acts Alone, but Few Can Govern It: Deloitte

Two in five companies now let AI execute decisions or actions without human approval to at least a moderate degree. Yet among those running multiple AI agents in one workflow, just 9% say the workflows are fully governed.

The findings are from Deloitte's The State of AI in the Enterprise report, published in October by the Deloitte AI Institute.

Deloitte surveyed 3,500 senior leaders across 25 countries and six industries between May and June 2026, including 100 in Australia. All participating companies had at least one working AI implementation in daily use.

Respondents were board and C-suite members and leaders at president, vice president and director level, split equally between IT and line-of-business roles. The survey was supplemented by 15 executive interviews.

Half of the companies running multi-agent workflows either have no formal governance or rely only on basic oversight.

Accountability is also patchy. When an AI system produces a negative outcome, 36% hold the team that owns it responsible.

Another 24% assign responsibility to a central governance function, and 10% to the employee who prompted the system. Nearly one in four (23%) say accountability is unclear.

Deloitte warns that when responsibility is unclear, problems can move between teams without anyone having the authority or obligation to resolve them.

The report argues oversight should match the authority given to AI. A system that recommends an action does not need the same oversight as one that carries it out.

Faster, Not Different

Companies need to decide which actions require human approval, which can be reviewed afterwards, and which should trigger immediate escalation, according to Deloitte.

AI is delivering speed more than change. Some 81% say it has improved the speed of decision-making and execution.

Only 25% are using AI for deep transformation, such as new products, reengineered core processes or new business models. That is down from 34% in Deloitte's January report.

Job redesign is similarly shallow. Some 60% have redesigned jobs around AI to at least a moderate degree, but only 15% describe the changes as extensive or complete.

Only 15% report a significant or fundamental shift in decision-making authority as a result of AI.

Cost visibility thins out below the headline bill. Some 71% track total AI inference and API spend, but only 51% track token volume and 44% track per-unit inference cost.

Some 86% expect AI investment to rise in the coming fiscal year, and 34% expect an increase of 20% or more.

Returns are improving. Revenue gains from AI are now reported by 28% of companies, up from 20% in January. Efficiency and productivity gains rose from 66% to 74%.

Sovereignty and Physical AI

Data residency and in-region compute are at least moderately important to the AI strategy of 74% of companies.

Some 62% are at least moderately concerned about reliance on foreign-owned AI technologies and compute.

Among those concerned, roughly half are prioritising vendors from preferred countries, building in-house capability, or shifting to on-premises or private cloud.

Physical AI, which combines AI with sensors, controls and robotics, is used at least moderately by 25% of companies. That share is expected to reach 48% within two years.

Asia Pacific is ahead, with 31% reporting at least moderate use today. Only 19% of all companies rate their Physical AI expertise as high or very high.

 

Two in five companies now let AI execute decisions or actions without human approval to at least a moderate degree. Yet among those running multiple AI agents in one workflow, just 9% say the workflows are fully governed.

The findings are from Deloitte's The State of AI in the Enterprise report, published in October by the Deloitte AI Institute.

Deloitte surveyed 3,500 senior leaders across 25 countries and six industries between May and June 2026, including 100 in Australia. All participating companies had at least one working AI implementation in daily use.

Respondents were board and C-suite members and leaders at president, vice president and director level, split equally between IT and line-of-business roles. The survey was supplemented by 15 executive interviews.

Half of the companies running multi-agent workflows either have no formal governance or rely only on basic oversight.

Accountability is also patchy. When an AI system produces a negative outcome, 36% hold the team that owns it responsible.

Another 24% assign responsibility to a central governance function, and 10% to the employee who prompted the system. Nearly one in four (23%) say accountability is unclear.

Deloitte warns that when responsibility is unclear, problems can move between teams without anyone having the authority or obligation to resolve them.

The report argues oversight should match the authority given to AI. A system that recommends an action does not need the same oversight as one that carries it out.

Faster, Not Different

Companies need to decide which actions require human approval, which can be reviewed afterwards, and which should trigger immediate escalation, according to Deloitte.

AI is delivering speed more than change. Some 81% say it has improved the speed of decision-making and execution.

Only 25% are using AI for deep transformation, such as new products, reengineered core processes or new business models. That is down from 34% in Deloitte's January report.

Job redesign is similarly shallow. Some 60% have redesigned jobs around AI to at least a moderate degree, but only 15% describe the changes as extensive or complete.

Only 15% report a significant or fundamental shift in decision-making authority as a result of AI.

Cost visibility thins out below the headline bill. Some 71% track total AI inference and API spend, but only 51% track token volume and 44% track per-unit inference cost.

Some 86% expect AI investment to rise in the coming fiscal year, and 34% expect an increase of 20% or more.

Returns are improving. Revenue gains from AI are now reported by 28% of companies, up from 20% in January. Efficiency and productivity gains rose from 66% to 74%.

Sovereignty and Physical AI

Data residency and in-region compute are at least moderately important to the AI strategy of 74% of companies.

Some 62% are at least moderately concerned about reliance on foreign-owned AI technologies and compute.

Among those concerned, roughly half are prioritising vendors from preferred countries, building in-house capability, or shifting to on-premises or private cloud.

Physical AI, which combines AI with sensors, controls and robotics, is used at least moderately by 25% of companies. That share is expected to reach 48% within two years.

Asia Pacific is ahead, with 31% reporting at least moderate use today. Only 19% of all companies rate their Physical AI expertise as high or very high.