AI governance — the rules and processes that guide how artificial intelligence systems are developed and used — is often seen as a barrier to progress, but when done well, it can support the responsible growth of AI technologies. A recent study by ABBYY on intelligent automation found that business leaders are split on who should be accountable when AI systems cause harm or make errors. Nearly half of French executives (44%) believe AI is being adopted faster than their companies can manage it effectively, highlighting the risk that safeguards are not keeping up with the pace of AI development, potentially leading to long-term compliance issues. When asked about responsibility for AI outcomes, responses were varied. About a quarter (24%) believe both the organization using AI and the provider should share responsibility, while 26% say the organization using the AI should be primarily accountable. Another 21% think the provider should be solely responsible, 16% place the onus on the end user, and 12% believe regulators should take the lead. This lack of consensus can lead to confusion, delays, and regulatory challenges when quick and clear action is needed. Clarity about responsibility also declines as you move down the organizational hierarchy. While 88% of senior executives in France know who is responsible for AI development and management, this drops to 84% among middle managers and only 73% among employees — those most directly involved in AI operations. This uncertainty can delay problem resolution and increase the risk of errors or non-compliance. Despite the growing use of AI, French executives show limited trust in the technology. Only 26% fully trust AI to produce accurate results, 24% trust it to explain important decisions or protect confidential information, and 22% trust it to avoid introducing unacceptable risks. However, 27% believe their AI systems are compliant with regulations. Many executives still see governance as a hurdle to innovation, with 67% saying data governance rules slow down AI deployment, and 26% citing ethical concerns as a significant delay. The biggest challenge for AI's return on investment is not the complexity of the AI model itself but the data it relies on and the trust in that data. Organizations must know where their data comes from, whether it is reliable, how it can be used, and where it is stored and processed. As data sovereignty — the control of data by the entity that owns it — becomes more important, there is a growing need for machine-readable controls that ensure consistent compliance throughout a data's lifecycle. Research shows that it is not governance itself that slows AI, but weak governance. Organizations that treat governance as a foundation rather than a barrier achieve better results: 72% of French executives say their AI initiatives have been more successful because of strong governance, and 70% believe their governance approach effectively balances risk with the need to deploy AI quickly. Although 77% of French organizations claim to have formal AI governance frameworks, fewer have practical mechanisms in place to handle AI-related incidents. Only 52% have an incident reporting process, 45% have an incident response plan, 34% have rollback capabilities, and 32% have an emergency stop button for AI systems. This gap between policy and preparedness leaves companies vulnerable to reputational and regulatory damage. Human oversight remains a critical safeguard. More than half (55%) of French executives say humans review all important AI decisions before any action is taken, though this varies by region, with 61% in Australia and 49% in Singapore. As AI systems make more high-stakes decisions, maintaining a human in the loop is not a sign of hesitation but a mark of responsible governance. The path forward does not require slowing down innovation but closing the gap between AI adoption and accountability. The most important steps are assigning responsibility clearly, strengthening the data foundation, and building safety mechanisms beyond just a framework. Organizations that answer the question "who is responsible for AI?" before problems arise will be those that can deploy AI confidently and effectively.