Two developments from 22 July that matter directly for Dutch organisations: OpenAI launched a new enterprise platform for autonomous AI agents, and the Digital Omnibus agreement of May 2026 postpones a series of EU AI Act deadlines, so the pressure around August 2026 is smaller than many teams assumed.
OpenAI Presence: agents for narrowly defined business processes
On 22 July 2026, OpenAI introduced Presence, an enterprise platform that lets businesses deploy AI agents for specific, clearly bounded workflows, from supplier onboarding to IT service desk. The platform is not self-service: implementations run through OpenAI Forward Deployed Engineers or selected systems integrators, and it is currently available to a small group of enterprises through a limited general availability programme.
What sets Presence apart from standard API use is its built-in policy layer: guardrails, approved actions per agent, a simulation tool to test safety measures, and escalation rules that route a conversation to a human employee when the agent steps outside its remit. OpenAI's own English-language telephone support already runs on it, 75% of incoming issues are resolved without human intervention.
GDPR consideration: Presence collects agent telemetry that is used for product improvement, including analysis of how often an agent escalates. Organisations that have Presence process customers' personal data need to be clear whether OpenAI acts as a processor and whether a data processing agreement is available. OpenAI has not yet published public EU-specific documentation for Presence. Caution is advisable until the data processing arrangements are transparent.
For context on the wider agentic AI tools market, see the model comparator and the safety page.
Digital Omnibus: AI Act deadlines shifted
Many organisations were preparing for August 2026 as the point at which high-risk AI obligations would fully apply. Under the Digital Omnibus agreement of 7 May 2026, those deadlines now fall differently:
| Obligation | Previously | Now |
|---|---|---|
| Annex III high-risk AI (standalone systems) | 2 Aug 2026 | 2 Dec 2027 |
| Annex I high-risk AI (regulated products) | 2 Aug 2027 | 2 Aug 2028 |
| Watermarking of AI-generated content | Aug 2026 | 2 Dec 2026 |
What has not shifted:
- Article 4 (AI literacy obligation): in force since 2 February 2025. Providers and deployers must ensure staff have sufficient knowledge of the AI systems they use.
- Article 5 (prohibited practices): likewise in force since 2 February 2025. Prohibits, among other things, social scoring, manipulative subliminal techniques and real-time biometric identification in public spaces.
The shift gives organisations more time to prepare technical documentation and work through conformity assessments for high-risk systems. But the core obligations, training staff and excluding prohibited uses, have already applied for a year and a half.
Dutch entrepreneurs and HR departments using AI for recruitment, credit assessments or medical support (all in Annex III) gain extra room from this postponement. More on the risk classes is available on the safety page.
What does this mean this week?
- Using Presence? Hold off on processing personal data until OpenAI publishes a data processing agreement for Presence and it is clear in which region data is processed.
- EU AI Act planning: If your high-risk system falls under Annex III, you have until December 2027. Use that time for a thorough conformity plan, not as a reason to delay the AI literacy obligation, which is already in force.
- Comparing? On the sector pages you can see, per sector, which tools and models offer relevant certification or EU hosting.