What actually changed?
The EU reached agreement on a set of amendments that push back the AI Act's most demanding deadlines. The obligations for high-risk AI systems, which many companies were racing to meet by August 2026, now fall in late 2027 and 2028 depending on the category. That is a meaningful shift, and it is being reported as relief for business.
What did not change?
The parts that reach the widest set of companies stayed where they were. Transparency duties still apply from August 2026: if you use AI to make or materially shape decisions about people, you have to be open about it. And from December 2026, AI-generated content has to be marked as such. So the deadline that moved is the one for a specific, smaller set of high-risk systems; the duties that touch ordinary, everyday use of AI did not move at all.
How should a serious company treat the delay?
As time, not as relief. A postponed deadline only helps the organizations that use the extra months to get their governance, documentation, and disclosure in order before the rules bind. The ones that read a delay as permission to wait will arrive at the new deadline with the same unfinished work and far less room to do it well. The disciplined move is to keep building the governance now, while it can be done calmly, and to treat August and December 2026 as the dates that already matter.
Read next: ISO 42001 Is Now a Procurement Requirement. Is Your AI Program Ready? · The AI Agent Governance Gap: Why Fast Deployment Is Building Tomorrow's Liability