Key pointSlide 01 / 15
Rules may be delayed. Responsibility cannot.
The regulatory calendar has shifted. Contractual, data protection, cybersecurity and management responsibility has not been put on hold.
Eventaisummitbudapest.hu
AI Summit Budapest 2026 · Future Lab · House of Music Hungary · 7 September, 09:20
Rules may be delayed. Responsibility cannot.
Two important AI Act dates have been pushed back. Meanwhile models are updated, data moves, dependence on vendors grows, and the company is taking decisions today that it must be able to defend later.
▶ View the talk · 15 slides
▶ Recording · 20:44
Five questions for executives
AI coffee: 30 minutes, free of charge
Counsel, Head of the IT, Data & Cyber practice · Taylor Wessing Budapest · The Legal 500: Leading individual 2025 (Taylor Wessing profile, opens in a new tab)
The full talk
The fifteen slides
Beside each slide you will find its main claim, the speaker’s outline, a practical explanation and sources you can check. The slides are shown in Hungarian, as presented.
← → browse · space play · swipe on mobile
Keywords5
Hover over a keyword or tap it: the outline lights up where it was said.
drkj.hu/en/aisummit/#dia-01
Speaker’s outline2 points
- My third year on this stage. In 2024 I spoke about the ecosystem, in 2025 about data. This year, about who is responsible.
- The question is not whether there is an AI strategy. It is whether a specific decision can be defended later.
In practice2 notes
The central question of these twenty minutes is simple. If one of the company’s AI decisions were challenged tomorrow, could we show which system version was running, on what data, with what access rights, who approved it and how it could be overridden?
An authority is not the only one that may ask. Customers, insurers and investors want the same answer, and so does the court in any later dispute.
Sources and data1 source
- Official programme: AI Summit Budapest 2026, Future Lab, 09:20 to 09:40. source
Two takeaways
Same AI. Two scores.
Two double-sided decision sheets for executives: one for before an AI product is released, one for before a company puts AI into use. Each has its own checklist with status boxes to mark. Those who build and deploy use both. Both sheets are in Hungarian.
For leaders who build AI products
Code to Market
Taking an AI product to market: legal role, classification and the responsibility you take on. An eight-point checklist for the release decision, with status boxes to mark.
↓ Download | PDF, 2 pages, in Hungarianv2.0 · 14 October 2026 · General professional information, not legal advice on any individual matter.
For leaders of companies deploying AI
Pilot to Control
Bringing AI into the company: actual use, vendor dependencies and operational control. An eight-point checklist for going live, with status boxes to mark.
↓ Download | PDF, 2 pages, in Hungarianv2.0 · 14 October 2026 · General professional information, not legal advice on any individual matter.
Next step
Bring one specific use case. Thirty minutes show the next step.
In the first conversation we take one specific AI system, procurement or development. Together we find the three questions that need an answer before the next decision.
Board Readiness
For executive decisions: what actually runs, who is responsible for it, and which control does the approval depend on?
Pilot to Control
For deployment and procurement: do the technical operation, the contract and the decision audit trail fit together?
Code to Market
For developers: is the product ready for the customer questionnaire, change management and market entry?
The packages in detail on the Advisory page
AI coffee at the Gerbeaud House
One real system takes us further than a general booklet.
A free orientation conversation at Taylor Wessing’s Budapest office; it does not by itself create a lawyer-client relationship. Details and terms: Advisory
The recording
Twenty minutes from the Future Lab stage.
The full recording of the talk (20:44, in Hungarian) was published on AI Summit Budapest’s official channel on Indaplay. One click opens it in a new tab.
▶Watch the recording (Indaplay, opens in a new tab)Indaplay · AI Summit Budapest 2026 · Future Lab · 20:44
The recording opens in a new tab on Indaplay, an external site, on AI Summit Budapest’s official channel; Indaplay’s own privacy and cookie rules apply there. This page does not embed an external player.
Five questions for executives
What would I ask before the next approval?
If any of them lacks a specific, documented answer, the use case is not yet ready for approval.
01What AI is actually running in the company?
+
Besides the approved systems, count the features built into existing software, individual use and agents that can act on their own.
02Is yesterday’s approval still valid?
+
A model swap, a new data source, a plug-in or a new access right can change the system so that the earlier review no longer covers the same operation.
03How far does the system’s mandate extend?
+
Technical access rights must not be broader than the business authorisation. It must be clear what the system may do on its own and where a genuine human decision is required.
04Did the vendor’s promise make it into the contract?
+
A provider calling its own operation compliant does not by itself protect the deploying company. Change notification, data boundaries, liability and exit must also appear as enforceable commitments.
05Could a contested decision be reconstructed within a day?
+
After the event, what counts is whether the model, the version, the data used, the control applied and the human decision can be identified. A log of the output is not always enough on its own.
Four numbers
What slipped, and what did not
The figures reflect the position on 7 September 2026. Detailed sources are given under the related slides.
825m
Uber fine in euros · GDPRDutch Data Protection Authority · 21 August 2026 · not final451 days
to the high-risk deadlineAnnex III · 2 December 2027 · Annex I · 2 August 20282,132
organisations audited on time | NIS2SZTFH | of 2,520 subject to the audit obligation | 10 July 202677%
where AI outpaces governanceIBM Institute for Business Value · June 2026 · 2,000 executives · 33 countriesA note on language
I gave this talk in Hungarian, on the Future Lab stage of AI Summit Budapest on 7 September 2026, and the slides on this page are shown as presented, in Hungarian. Their claims, outlines, practical notes and sources are given here in English; the recording and the PDFs remain in Hungarian. The original page is in Hungarian at drkj.hu/aisummit.
The material of the talk is also available as in-house training, with your organisation’s own questions. Request in-house training
Content. This page provides general professional information and is not legal advice on any individual matter. References and deadlines reflect the position on 7 September 2026.
Use of AI. I also used AI-based tools to prepare the talk materials and this page. The texts and images went through human professional review and editing, and I bear editorial and professional responsibility for the content; for the texts, this is the human review and editorial responsibility referred to in the second subparagraph of Article 50(4) of the AI Act.
Reuse. Short quotations may be used with attribution. Republishing the full material, or any self-contained part of it, requires prior consent.
Design and recording. The colours and typography of this page refer to the visual identity of AI Summit Budapest 2026; the event was organised by Economx and IndaEvents, and the recording is published by Indaplay. This is my own professional page, independent of the organiser.