My flight landed in Budapest eighteen hours before the talk. This was enough time to review my notes, go for a walk to check on the local scene and try to get some sleep despite the struggle with the time zone change. I left the hotel, turned left and headed to Szabadság-szobor, the Liberty Statue, nestled on top of a hill overlooking the Danube River. The monument originally celebrated the city’s liberation from the Nazi occupation during the Second World War with an austere style that echoes the years of Soviet domination behind the Iron Curtain. The views were magnificent. The sun was setting behind the hilly Buda side, casting beautiful yellow shades over Pest. Buda and Pest, two sides of a city separated by the river. A perfect metaphor for AI deployment: the duality of speed and progress contrasted with responsible and ethical use. How trust could reconcile both was going to be at the center of my talk the next morning.

Enterprise AI adoption is accelerating. The International Energy Agency estimated that major technology companies’ capital spending would rise 75% in 2026 to approximately $715 billion. Its satellite tracking found that AI data center capacity more than tripled in the eighteen months until April 2026. Inside companies, adoption is intensifying. In August 2026, OpenAI reported that its top enterprise customers generated 8.3 times as much output per active user as typical customers.

The way companies use AI is also changing. At first, the system supported individual tasks, such as summarizing texts or drafting memos, and a person stayed close to the work. Users then moved to using the technology to write code and, in parallel, embed AI into business workflows for decision support. Today, AI agents complete complex tasks and, with each step, more authority gets delegated.

Recognize The Competing Pressures During AI Deployment

Since last year, I have argued here in my Forbes column about AI’s three limits : economic, physical and moral, which more recently I have cast as institutional given the direct linkage to regulations. Businesses must earn returns for their AI investments and the business models of the AI firms themselves need to hold water. The infrastructure upon which AI relies, notably the data centers that power the execution of AI models, depends on chips, electricity, water and land. Even a profitable and feasible use can still weaken rights or remove human judgment from decisions that deserve deliberation.

An ethical lens helps make these choices explicit, especially when under uncertainty. When deploying AI, leaders are the ones who should determine which judgments and decisions should remain in human hands. Equally, they are the ones who should assess the impact of the systems being implemented and determine what uses of the data are acceptable, who may be disproportionately impacted by errors and who receives benefits while others bear resource costs. These responsibilities, including overall system safety and security, span the company’s value chain.

Enterprises find themselves squeezed between two forces. Commercial pressure favors pursuing greater market share and increasing delegation of decisions to automation for efficiency. On the other hand, responsible deployment requires detailed testing in context and preserving human authority to correct or stop a system when needed.

Build Trust Into AI Deployment

Without trust, it will be difficult to sustain the current boom in AI enterprise demand. Trust can be the connecting glue between the two conflicting forces we just discussed. However, in this context, trust assumes a very specific operational meaning. It requires evidence that AI works in its actual setting, clear named ownership for when it fails and accountability for its effects. Inside the enterprise, employees need to know when they can intervene and, outside, customers need a credible path to correction. On top of all, company boards need documented evidence that management is ready and able to respond and adapt when conditions change.

As an example, let’s look at a manufacturer using AI to assign overtime shifts. The system may improve production coverage while repeatedly burdening workers with caregiving commitments. The process should account for human intervention, allowing a plant manager to examine the assignments and preserving human authority to change the scheduling rules. Workers need a way to challenge an assignment before it takes effect, without risking their jobs–elements of process design and company culture, exogenous to the AI system. Trust grows when employees can see that raising a concern leads to a fair review and an actual correction.

Ethical governance works only when each part of the business knows what it owns and in what context. Executives set the limits and escalation criteria for when an issue should go up the chain. With the boundaries defined, each function plays a role. Product and portfolio managers keep inventory of AI uses and sort them by how much impact a failure could cause. The engineering team tests the systems in context and end-to-end since individually working parts can still fail in combination. People and process leaders inspect the workflows to ensure that reviewers have enough time and training to do their jobs and real standing to disagree with a system recommendation. At one end, procurement teams ask vendors for evidence of performance, require notice before changes and negotiate a way out if the relationship ends. At the other, assurance teams follow the results over time and check whether identified problems lead to fixes and organizational learning.

We can learn lessons from this tension between commercial acceleration and responsible deployment. Value and return on investment calculations must include the cost of making systems act responsibly and workflows must allow for human intervention. Bear in mind that we assign permissions to the systems that we create . AI governance, as any governance, depends on human judgment and organizations should create a culture that rewards employees for exercising it.

The bridges of Budapest join the two sides of the city. Ethical reflection can help build the trust to do the same for AI deployment, helping managers pursue commercial goals responsibly.