On 28th July 2026, more than 1,100 people working inside the world's leading AI companies signed a letter asking governments to build the ability to slow frontier AI down, if it ever becomes necessary. Here's what that signals for regulation, disclosure, and the far more mundane problem of ad hoc AI creeping into everyday business processes.
Originally published on LinkedIn. Read the original
On the 28th July 2026, something happened that we should all take notice of.
More than 1,100 people working at some of the world's leading AI companies signed a letter asking the US government to support a global initiative to develop the mechanisms needed to slow down the development of frontier AI, if it becomes necessary.
They weren't asking for AI development to stop today. Their point was more subtle, and perhaps more important: if things start moving too quickly, society may need the ability to buy time. Left purely to commercial and geopolitical pressures, no individual company or country is likely to volunteer to slow down while everyone else carries on at full pace.
Note, this was not an anti-AI group asking for this. It was people inside OpenAI, Anthropic, Google, Meta and others, including some extremely senior people who are actually building these systems — companies into which billions have been invested, with enormous commercial incentives to keep moving quickly.
This is a big deal.
It doesn't prove that AI is about to undergo what is sometimes called a 'take-off', but the reason given for the concern is fascinating. The letter says the leading AI companies believe they could be close to automating AI research itself.
One possible consequence of that is something called Recursive Self Improvement (RSI). Put simply, AI becomes increasingly involved in creating better AI. If each generation becomes better at helping to create the next generation, you potentially create a feedback loop. If that loop accelerates sufficiently, it could result in what is known as a fast take-off or intelligence explosion.
That could eventually produce broadly human-level Artificial General Intelligence (AGI), followed by systems significantly more capable than humans — usually referred to as Artificial Superintelligence (ASI). The more speculative idea of a technological singularity describes a point beyond which technological progress becomes so rapid and profound that predicting what happens next becomes extremely difficult.
None of this is guaranteed. There are researchers who believe automated AI research could result in dramatic acceleration, and others who believe practical limits — compute, energy, economics, algorithms and the real world — will prevent anything resembling runaway self-improvement.
We simply don't know.
But if it does happen, there is a possibility that the resulting systems become harder for humans to understand and ultimately harder to control. Hence the desire to build the brake before anyone actually needs to press it.
One thing is certain: AI is changing the world, and this will continue at pace.
There will be regulation. There may well eventually be new taxes or levies. Governments are already demonstrating that they are prepared to intervene directly. In June, the US government actually imposed export controls on Anthropic's newest Fable 5 and Mythos 5 models, preventing access by foreign nationals. Those particular restrictions were lifted a couple of weeks later, but the precedent is interesting.
I would not be surprised to see a move towards some form of global AI Inspectorate, perhaps similar in concept to the International Atomic Energy Agency. In fact, the idea isn't particularly new: OpenAI itself has previously suggested that sufficiently powerful AI might eventually need exactly that sort of international oversight.
That might sound dramatic, but sufficiently powerful AI has potential consequences for national security, critical infrastructure, cyber security and society that governments simply aren't going to ignore.
I also think it is extremely likely that disclosure of where and how AI is being used by companies and public services becomes increasingly mandatory.
That process has already started.
UK government departments already have mandatory transparency requirements for certain algorithmic systems, and the EU AI Act now contains explicit transparency requirements around particular uses of AI. We aren't yet at the point where every company has to publish an 'AI register', but I wouldn't bet against us getting there.
And this is where things become rather more mundane — but possibly more relevant to most businesses.
The race to take first-mover advantage has resulted in many IT and operational teams embedding AI into processes and software in an ad hoc, spider-web manner. It works. The business is pleased. Another API gets added. Another prompt. Another integration.
Then one day something goes wrong.
Or an auditor starts asking questions.
The person who created it has left. The API it connects to changes. Nobody is quite sure which model is making which decision or why. The problems start.
Many companies are now starting to think properly about AI policies to address this. They are talking about transformation and how the use of AI can break the traditional Quality, Speed, Price iron triangle — which it can! — but they need to do it in a controlled way.
They need to identify where the human touch is sacrosanct, where it is simply habit, and indeed where it is destructive and wasteful.
Understanding processes is where it begins. Once you can see what you are doing today, you can start to work towards what you want to do tomorrow.
Using a process platform such as Opzaro, Camunda, ProcessMaker or FlowForma, to name a few, allows the business process to remain the thing that drives the technology, rather than letting the technology dictate the process. The things that give the business its USP remain front and centre.
Systems such as Opzaro take this a stage further by making AI a first-class participant in the process, but giving it a defined job. Its role is visible. Its actions are auditable. A business person can see where AI is being used and understand the prompts, inputs and outputs involved.
I think this is going to become increasingly important.
AI adoption isn't going to stop. Nor should it.
The challenge is going to be making sure that, five years from now, we can still explain what we built.
Disclosure: the author is directly affiliated with Opzaro.
Opzaro gives AI a defined, auditable role inside a governed process rather than a hidden prompt bolted onto a script. You can see where AI is being used, what it was given, what it decided, and who is accountable — today, and whenever the regulator or the auditor eventually asks.
That's the foundation you need before disclosure requirements and AI registers stop being optional.
Opzaro coordinates APIs, AI models, business rules and human tasks inside one governed, auditable workflow engine.
Explore AI & automationWe'll show you how Opzaro makes AI use visible and auditable inside your own workflows — not just slides.