Who you'd meet
Both founders. There is no sales engineer to hand you off to.
It is the one question every AI policy implies and almost no organisation can answer. Not what the model was allowed to do. Not what the framework says it should do. What it did, on Tuesday, at 14:20, to a real customer record.
You would meet both founders. There is no sales engineer to hand you off to.
EU hosted · your own keys · runs on your infrastructure
"Write me a draft message. Do not send it."
What happened next: the assistant opened LinkedIn in a browser, composed the message to a named executive at another company, and began to send it. A human noticed and stopped it by hand, mid-sentence, during a call.
This was not a customer. It was one of us, on our own machine, with more monitoring running than most enterprises have. And here is the part that should interest you: it left no trace anywhere. Every policy we had was written around the model, what it may be asked, what it may answer. The action went out through a browser window, which no policy was watching.
Nothing in a governance framework would have caught that. Not because the framework was bad, but because the framework governs a document, and the agent was operating a machine.
Policy is written. Agents act.
The distance between what your policy says and what your agents did is currently unmeasured. In most organisations it is also unmeasurable.
Almost all AI governance work today produces artefacts: a policy, a register, an impact assessment, a training slide. They are genuinely necessary and they are all descriptions of intent. Meanwhile the thing being governed has quietly changed shape. It is no longer a model answering questions. It is software with hands. It opens systems, writes records, sends things to people, and spends money, thousands of times a day, mostly unobserved.
That is not a criticism of the profession. It is a tooling gap, and it is roughly where financial controls were before anyone thought to keep a ledger. The interesting work of the next two years is not writing better policies. It is being able to show, afterwards, what happened. That is the plain version of what we build, and the whitepaper is the long one.
The Digital Omnibus entered into force on 27 July 2026 and moved the high-risk timetable. Everything here is public, dated, and linked at the bottom.
| Obligation | Applies from | Where that leaves you |
|---|---|---|
| Banned uses, and the duty to make staff AI-literate | 2 Feb 2025 | In force |
| Rules for general-purpose model providers | 2 Aug 2025 | Your vendors' problem |
| The Act generally, including telling people they are dealing with AI | 2 Aug 2026 | In force |
| High-risk uses under Annex III: biometrics, critical infrastructure, education, hiring, essential services, law enforcement, migration and borders | 2 Dec 2027 | Moved back by the Omnibus |
| High-risk built into regulated products under Annex I | 2 Aug 2028 | Moved back by the Omnibus |
Sources, read 1 September 2026: European Commission, Regulatory framework for AI; Regulation (EU) 2024/1689 (the AI Act), Official Journal; the Digital Omnibus amending Regulation (EU) 2024/1689, European Commission.
We will say this plainly because we would rather be argued with than agreed with politely.
In the conversations we are having across Europe right now, law firms, payroll companies, industrial groups, governance is never the reason the meeting happens. The reason is always the same: someone has looked at what their teams are spending on AI, and at how fast that number is moving.
One managing partner walked us through a €150,000 offer he turned down, not on price but because he could not see where the price would be in a year. A German security executive got there from the opposite direction and reached the same place. In both rooms, control and evidence were things they wanted, but they were not what got the meeting.
"We need to be compliant with the AI Act, and we need a policy for agents."
"I cannot see what this is costing me, or what it is doing, and I have to sign for both."
Once the spend is visible, the same feed answers the governance question. It is the same record.
We show you what real agent activity looks like inside a company, the actions, the near-misses, the spend nobody attributed to anyone, including what our own policy missed, and you tell us where our reading of this market is wrong.
Both founders. There is no sales engineer to hand you off to.
Real observed activity, and the questions we haven't answered.
Your read on where this market actually is.
European hosting, your own keys, runs on your infrastructure.
TapPass.ai · the authority layer for agentic AI