On 27 September 2026, Reuters reported that Donald Trump had confirmed plans for a dinner with Anthropic chief executive Dario Amodei. The planned meeting put competing views on AI development in the same room. For businesses, it raises a familiar issue: important conversations need a path into decisions that survive beyond the conversation.

What happened

Reuters reported that Trump continued to oppose slowing AI development, emphasising competition with China, while acknowledging Amodei’s concerns about moving too quickly. The report established the planned discussion and the disagreement. It did not establish a resulting agreement or a new governance mechanism.

Personal access can create an opportunity to explain a risk. It cannot, by itself, tell an organisation which evidence is sufficient, who may authorise deployment or what happens when a previously approved system behaves differently. Those questions require a process that other people can understand and use.

Why it matters

Imagine a division head securing the chief executive’s verbal support for a customer-facing AI assistant. The sponsor leaves believing the project has permission to launch. Security believes only a pilot was approved. Operations assumes that a human will review every response. No one is deliberately obstructing progress, but the same conversation has produced three different operating assumptions.

A concise decision record could prevent much of that confusion. It should state the approved use, the limits, the evidence considered and the person responsible for unresolved conditions. It should also say what has not yet been decided. That last point matters when enthusiasm for the overall direction is mistaken for approval of every implementation detail.

This is not a demand for a large committee around every experiment. A contained internal test can use a lighter process than an assistant that commits the company to customers. The discipline is to make the level of authority match the consequence of the action.

The bigger shift

AI decisions can change as the system, its access and its users change. Approval for a drafting assistant should not silently become approval for an agent that sends messages or changes records. The original decision needs a boundary, and material changes need a route back to the people authorised to assess them.

That is why a practical AI governance framework must connect leadership intent with daily operations. A policy document can express priorities, but somebody still needs to answer a release question, resolve a disagreement and support a pause when evidence is incomplete.

Useful governance also preserves the reasons behind a decision. If the original sponsor moves roles, the team should not have to reconstruct permission from memories of a meeting. A recorded rationale allows the next owner to see which assumptions remain valid and which deserve another look.

My take

Senior conversations are valuable when they lead to clear responsibilities. They become fragile when they are treated as the whole control system.

At the next AI steering discussion, select one live decision and write its conditions in plain language. Name the decision-maker, the operating owner and the person who can stop the activity. Specify what evidence would trigger reconsideration. Keep the record short enough that the people running the process will actually read it.

Then test the arrangement with a plausible change: the assistant needs a new data source, begins sending external messages or produces an unexplained result. If the team knows where to take that change, leadership intent has become operational. That is a more durable achievement than leaving a meeting with general agreement that AI is important.