The AI Governance Gap: Why Growing Confidence Is Outpacing Real Accountability

Ask 10 business leaders who should be held responsible when an AI system produces a harmful or incorrect output, and you will get at least five different answers. That is not a hypothetical. It is exactly what happened when ABBYY surveyed 1,200 senior managers across six countries for the 2026 ABBYY State of Intelligent Automation report, Who Answers for AI: The Governance Gap.

This finding is worth sitting with for a moment. AI is now embedded in business-critical processes across enterprises worldwide. AI is influencing hiring decisions, credit approvals, customer interactions, and supply chain forecasts. And yet, the people deploying it cannot agree on a fundamental question: when the technology fails, who is accountable?

This is the governance gap. It is not a gap in ambition or investment. It is a gap in clarity, oversight, and trust, and it is widening faster than most organizations are prepared to admit.

AI is outpacing the ability to govern it

The core tension in the data is simple. Adoption is accelerating. Governance is not.

More than half of business leaders, 54%, say AI is being adopted faster than their organization can effectively govern it. That single statistic reframes much of the current conversation about AI. The urgent risk is not that businesses are too cautious. It is that they are moving quickly while the guardrails lag behind.

This creates a compliance exposure that compounds over time. Every new model, workflow, or automated decision added without corresponding oversight increases the surface area for something to go wrong, and decreases the odds that anyone will know who is meant to fix it.

Nobody agrees on who is accountable for AI

Here is where the governance gap becomes most visible. When asked who should bear responsibility when AI produces harmful or incorrect outputs, business leaders splintered:

• 31% believe responsibility should be shared between the organization and the vendor.

   

• 26% say the organization using the AI should bear primary responsibility.

   

• 20% place it fully with the AI vendor.

   

• 12% point to the end user.

   

• 10% believe regulators should take the lead.

   

There is no majority view. No dominant consensus. Just fragmentation.

Fragmented accountability is dangerous for a straightforward reason: when responsibility is distributed across five possible parties, it effectively belongs to no one. In the event of a serious AI failure, that ambiguity turns into finger-pointing, delayed response, and regulatory exposure precisely when speed and clarity matter most.

The problem deepens inside organizations, too. Knowledge of who is responsible for AI drops sharply the further down the org chart you look. Leaders report that 89% of senior leadership knows who owns AI development, implementation, and management. That falls to 84% at middle management, and just 65% among junior employees, the very people often closest to how these systems operate day to day.

A junior employee encounters a critical issue in an AI system, but confusion arises because only 65% of their peers are aware of who is responsible for managing it. This lack of clarity delays resolution and increases operational risks.

Meanwhile, senior leadership, with 89% familiarity regarding AI ownership, remains far removed from the day-to-day challenges, creating a disconnect between strategic decisions and practical implementation.

The trust deficit behind the numbers

The accountability confusion sits on top of a broader trust problem. Business leaders are not yet confident in the systems they are deploying:

• Only 26% completely trust AI systems to produce accurate outputs.
• Only 26% completely trust AI to explain how important decisions were reached, or to protect confidential information.
• Just 22% completely trust AI to operate without introducing unacceptable risks.

These are strikingly low numbers for a technology already operating inside core business processes. They reveal an uncomfortable reality: organizations are scaling tools they do not fully trust, without a clear agreement on who answers when that trust proves misplaced.

Governance is being treated as a brake, not an engine

One reason the gap persists is a widespread misconception that governance and innovation are opposing forces.

70% of business leaders say data governance requirements, the controls that keep data accurate, secure, and compliant, are slowing AI deployment. A further 34% say ethical concerns considerably slow their AI initiatives.

Governance, in other words, is frequently experienced as friction.

That friction has consequences.

Less than half of leaders, 47%, say their AI initiatives have exceeded expected targets, and 15% say performance is falling below expectations. When we asked what stands in the way of stronger returns, the answers were revealing:

• Data quality was the top barrier, cited by 20% of leaders.
• Integration with existing systems followed at 15% .
• Implementation costs came in third at 13% .

Notice what tops that list. The biggest obstacle to AI ROI is not the sophistication of the model. It is the data and trust underneath it.

AI requires organizations to know where data comes from, whether it can be trusted, how it can be used, and where it is stored and processed. As data sovereignty grows in importance, a shift toward machine-readable controls that govern data throughout its lifecycle is required to ensure consistent compliance.

The counterintuitive truth: Governance is the key to AI success

Here is the argument I want to make plainly, because the data supports it: governance is not the thing slowing AI down. Weak governance is.

The organizations that treat governance as a foundation, rather than a hurdle, are the ones seeing better results:

• 78% of leaders say governance has made their AI initiatives more successful.
• 75% say their governance approach effectively balances risk mitigation with the need to deliver AI quickly.
• Governance frameworks make it easier to develop AI systems (57%) , scale beyond the pilot stage (57%) , and encourage employee adoption (59%) .

Read those numbers together and the picture inverts. Governance done well does not trade speed for safety. It enables both. It gives teams the confidence to move AI out of pilot purgatory and into production, because the controls, accountability, and oversight are already in place.

Having rules is not the same as being ready

There is a critical distinction hiding in the data between having a governance policy and being prepared for a governance failure.

On paper, most organizations look covered. 82% report having a formal AI governance framework in place. But the fail-safes tell a different story:

• Only 50%have an AI incident reporting process.
• Only 46% have an AI incident response plan.
• Just 35% have an AI kill switch.
• Just 35% have AI rollback capabilities.

In other words, a large share of organizations have written the rules but have not built the emergency brake. Having a framework governs how AI is supposed to behave. Incident response, kill switches, and rollback capabilities are safety mechanisms that determine what happens when it does not. The gap between those two is where reputational and regulatory damage lives.

Human oversight remains a vital safeguard.

56% of leaders say humans review all important AI decisions before action is taken, though the approach varies by market, rising to 61% in Australia and falling to 49% in Singapore. As AI takes on more consequential decisions, keeping a human in the loop is not a sign of immaturity. It is a mark of discipline.

Human oversight validates AI results, ensures reliability in high-stakes tasks, and supports continuous learning and model optimization for adapting to business needs. Integrating human reviewers at critical decision points builds trust in AI-driven workflows.

Closing the gap: What leaders should do now

The path forward does not require slowing down innovation. It requires closing the distance between adoption and accountability. Three moves matter most:

1. Assign accountability explicitly. End the ambiguity. Decide, in writing, who owns AI outcomes across the organization and the vendor relationship, and communicate it clearly at every level, from senior leadership down to the junior employees closest to the systems.
2. Fix the data foundation. Data quality is the top barrier to ROI for a reason. Machine-readable controls that govern data across its full lifecycle turn compliance from a bottleneck into a built-in feature.
3. Build the fail-safes, not just the framework. Move beyond policy documents. Invest in incident reporting, response plans, kill switches, and rollback capabilities before an incident forces the issue.

The organizations that answer the question "who answers for AI?" before something goes wrong will be the ones that scale AI with confidence. Those that leave it unanswered are simply deferring a reckoning.

The governance gap is closeable. But it will not close on its own.

ABBYY is a strong supporter of making documents used by AI compliant through its cooperation with the new document AI standard, DocLang.

We are also the first to embark on full support of BSI C5, a cloud security assurance framework specifically designed to evaluate the security and operational transparency of cloud service providers. Unlike broader security certifications that focus primarily on governance or management systems, BSI C5 was created to address the unique risks associated with cloud computing environments.

We are ready to help make your enterprise data ready for any AI system.

Explore the full 2026 ABBYY Who Answers for AI: The Governance Gap report for the complete data, market-by-market breakdowns, and a deeper look at how leading organizations are turning governance into a competitive advantage.