Aug. 2, 2026 arrived in Brussels on schedule12. Europe's AI law crossed its second big deadline. Chatbots now have to declare themselves. Images and audio changed by AI now have to carry a label. The European Commission's AI Office gained fresh power to enforce both. That date mattered less than its drafters once intended. A proposal from Nov. 19, 2025 had already moved the calendar3. It pushed the Act's toughest high-risk duties out to December 2027 and August 2028. Twenty-one companies had signed the voluntary code meant to ease that shift4. Two of the most watched labs in the field signed differently, or skipped the signature.
A Clock Set in Stages
The Artificial Intelligence Act entered into force on Aug. 1, 20241. Its authors built compliance in stages. Bans landed first, on Feb. 2, 2025, on systems Brussels judged too risky to allow. Social scoring made the list, along with hidden tricks that steer a user and some forms of biometric sorting. Rules for general-purpose models followed six months later. Providers had to publish technical documents and respect copyright. Most other operator duties, transparency among them, arrived at the two-year mark. High-risk requirements, the heaviest part of the Act, sat near the three-year mark. The Commission moved that one itself3.
Staggered deadlines gave companies time. They also gave regulators time to hire the staff needed to enforce a law longer than several national constitutions put together. Brussels used that time differently for different duties. Voluntary where officials hoped goodwill would carry, mandatory where the stakes justified a harder line.
Twenty-One Signatures, Two Exceptions
The Commission published its General-Purpose AI Code of Practice on July 10, 20254. Three chapters cover transparency, copyright, and safety. A Signatory Taskforce, chaired by the AI Office, runs it. Twenty-one companies signed. The list holds Amazon, Anthropic, Google, IBM, Microsoft, Mistral AI and OpenAI, plus smaller labs such as Aleph Alpha, Cohere and Fastweb. Microsoft framed its part as continuity5. Chief Responsible AI Officer Natasha Crampton wrote on the company's EU policy blog that Microsoft planned to build the Act's rules into how it makes products. She posted that nearly six months before the code existed.
xAI took a narrower path, signing the Safety and Security chapter and leaving transparency and copyright aside4. Meta stands apart. Its name is absent from the AI Office's signatory roster, which is hard to miss given the company's scale. A voluntary code rewards a company willing to trade paperwork for goodwill. Holding out is a bet that lawsuits, lobbying or delay cost less than compliance in every place the code might set the template.
The Omnibus Rewrites the Calendar
Brussels revisited its own timetable on Nov. 19, 2025, proposing a Digital Omnibus package with a dedicated AI provision inside it3. That provision entered into force July 27, 2026, six days ahead of the Act's next deadline. It rewrote the hardest parts of the schedule. Annex III covers hiring tools, credit scoring and decisions of that weight. Those systems now face duties starting Dec. 2, 2027, more than a year later than industry groups had planned for. Physical products with embedded high-risk AI, covered under Annex I, got until Aug. 2, 2028.
Aug. 2, 2026 brought transparency duties, and the Omnibus moved high-risk duties to December 2027 and August 2028
Dates in the EU AI Act's staged schedule, from entry into force in 2024 to the last high-risk deadline in 2028
Each pin is one date on a rail spaced by real time: blue pins are dates the Act itself set, and magenta pins are the Omnibus and the high-risk dates it moved.
Sources: Wikipedia [1]; European Commission, Digital Strategy [2]; European Commission, Digital Strategy [3]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Date | Event |
|---|---|
| Aug. 1, 2024 | Act enters into force. The Act enters into force |
| Feb. 2, 2025 | Bans apply. Bans on the riskiest systems apply; Social scoring, hidden tricks that steer a user, some biometric sorting |
| Nov. 19, 2025 | Omnibus proposed. Brussels proposes the Digital Omnibus |
| July 27, 2026 | Omnibus in force. The AI Omnibus enters into force; Six days ahead of the Act's next deadline |
| Aug. 2, 2026 | Transparency duties. Transparency duties take effect; Chatbots declare themselves, and AI-altered images and audio carry a label |
| Dec. 2, 2027 | High-risk duties. Annex III high-risk duties begin; Hiring tools, credit scoring and decisions of that weight |
| Aug. 2, 2028 | Product duties. Annex I product duties begin; Physical products with embedded high-risk AI |
Simplification came packaged with the delay. The Omnibus gave small mid-cap firms the lighter duties, widened the test sandboxes, added one at EU level, and cut the database paperwork for exempt systems3. Legislators also wrote a fresh ban, on AI built to make sexual or child-abuse material of a person who withheld consent. They widened the AI Office's reach over general-purpose models and over AI inside large online platforms. Read as strategy, the package trades near-term compliance friction for more time on the categories industry lobbied hardest against.
What Aug. 2 Actually Delivered
Duties that survived the rewrite proved narrower than the phrase "high-risk deadline" suggested. The Commission stated the first one directly: "Chatbots and other interactive AI systems will have to tell users they are dealing with AI, not a human."2 Labeling came next. "Deepfakes (images, videos, or audio that have been edited or generated using AI) will have to be labelled. AI-generated or altered content will also have to carry machine-readable marks so it can be detected more easily." Enforcement falls to the AI Office, working alongside national authorities in each member state.
Fines under the Act carry weight whichever duties apply on a given date. Violations of Article 5's prohibited practices draw fines up to €35 million or 7 percent of global annual turnover, whichever runs higher1. Other operator breaches draw up to €15 million or 3 percent. Misleading a regulator draws up to €7.5 million or 1 percent. Put plainly, one breach can cost more than a year of profit at a mid-sized firm. A ceiling that high holds a general counsel's attention on rules still years away.
A transparency breach can cost up to 3 percent of global turnover, against 7 percent for a banned practice
Maximum fine under the EU AI Act by type of breach, percent of global annual turnover
Each bar is one tier of fines: the magenta bar is the tier that covers transparency duties, and the gray bars are the tiers above and below it.
Source: Wikipedia [1]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Item | Value | Note |
|---|---|---|
| Banned practices, Article 5 | 7% | Up to €35 million or 7 percent, whichever runs higher |
| Other operator breaches | 3% | Up to €15 million or 3 percent; transparency duties fall in this tier |
| Misleading a regulator | 1% | Up to €7.5 million or 1 percent |
Brussels Hires, Labs Hire Back
Staff grew alongside the law they enforce. The AI Office now counts more than 125 people across six units6. Those units cover AI and robotics, rules and compliance, AI safety, policy, public good, and health. A hiring round for roughly 40 more agents closes Sept. 8, 2026. Brussels wants tech specialists, lawyers, operations staff and paralegals. Henna Virkkunen, an executive vice-president, holds the portfolio for tech sovereignty, security and democracy7. An Apply AI Strategy and a proposed European AI Research Council both come from that seat.
Labs staffed their own answer. Anthropic named Mariano-Florentino Cuéllar its first chief global affairs officer on Aug. 4, 20268. His brief covers Europe and the rest of the world. The hire landed during separate friction with Washington over export controls and a Pentagon supply-chain label. Read it as an admission. One Washington-facing policy chief covers too much ground for a company working under EU duties, U.S. export rules and fifty state legislatures at once. Crampton's post, months ahead of the code, shows the same instinct earlier: treat Brussels as a standing job, staffed years before the deadline that justifies it.
Industry's Relief, Parliament's Alarm
Industry treated the Digital Omnibus as a correction regulators owed them. Alexandre Roure, head of policy for CCIA Europe, reacted the day the Commission published the package, Nov. 19, 20259. "While the Commission has made progress on AI and should be commended for acknowledging the damage of regulatory overreach, significant gaps in the AI Act remain," Roure said. Roure thanked the Commission for one change while listing the AI Act provisions CCIA Europe still wants removed.
Parliament read the same announcement differently. Kim van Sparrentak, a Dutch member of the European Parliament, responded the same day10. "Disappointing to see the European Commission cave under the pressure of the Trump administration and Big Tech lobbies," she said. Her verb, cave, put the decision inside the Commission itself, an executive body she argued had buckled under outside pressure.
By the numbers
- €35 million or 7 percent of global turnover: the ceiling for a banned-practice breach under Article 5, whichever runs higher1.
- Twenty-one companies signed the Commission's General-Purpose AI Code of Practice, per the AI Office's mid-2026 roster4.
- Six months separated the Act's entry into force from its first bans, effective Feb. 2, 20251.
- July 27, 2026 marked the AI Omnibus's entry into force, eight months after its Nov. 19, 2025 proposal3.
- December 2027 is the new compliance date for Annex III high-risk systems, moved from the earlier August 2026 target3.
- Over 125 staff already work across the AI Office's six units, with roughly 40 more roles in active recruitment6.
What to watch
Sept. 8, 2026 closes the AI Office's current hiring window6. The roles it fills will show whether Brussels wants real power to inspect general-purpose models, or enough staff to move paper. Annex III's December 2027 deadline hands high-risk providers roughly fifteen extra months3. Watch what health care, hiring and credit vendors announce first. Meta's continued absence from the Code of Practice roster deserves another look whenever the AI Office refreshes it, since a later signature would reverse the company's current stance4. Anthropic's Cuéllar hire, set beside Microsoft's early Brussels spending through Crampton, points at a pattern58. European policy is becoming its own job at frontier labs, rather than an add-on to a Washington team.
Sources
- Wikipedia contributors, "Artificial Intelligence Act," Wikipedia, retrieved Sept. 4, 2026, https://en.wikipedia.org/wiki/Artificial_Intelligence_Act
- European Commission, "Commission Starts Enforcing AI Act Rules and New Transparency Requirements From 2 August," Digital Strategy, July 31, 2026, https://digital-strategy.ec.europa.eu/en/news/commission-starts-enforcing-ai-act-rules-and-new-transparency-requirements-2-august
- European Commission, "AI Omnibus Enters Into Force," Digital Strategy, July 27, 2026, https://digital-strategy.ec.europa.eu/en/news/ai-omnibus-enters-force
- European Commission, "Code of Practice for General-Purpose AI," Digital Strategy, July 31, 2026, https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai
- Natasha Crampton, "Innovating in Line With the European Union's AI Act," Microsoft EU Policy Blog, Jan. 15, 2025, https://blogs.microsoft.com/eupolicy/
- European Commission, "AI Office," Digital Strategy, Aug. 1, 2026, https://digital-strategy.ec.europa.eu/en/policies/ai-office
- European Commission, "Henna Virkkunen," College of Commissioners, June 24, 2026, https://commission.europa.eu/about/organisation/college-commissioners/henna-virkkunen_en
- CNBC Staff, "Anthropic Names Global Affairs Chief as Trump Tensions Persist," CNBC, Aug. 4, 2026, https://www.cnbc.com/2026/08/04/anthropic-names-global-affairs-chief-as-trump-tensions-persist.html
- CCIA Europe, "Digital Omnibus: Simplification of EU Tech Rules Requires Bolder Action," CCIA Europe, Nov. 19, 2025, https://ccianet.org/news/2025/11/digital-omnibus-simplification-of-eu-tech-rules-requires-bolder-action/
- Masha Borak, "EU Publishes Digital Omnibus Leaving AI Act Future Uncertain," Biometric Update, Nov. 19, 2025, https://www.biometricupdate.com/202511/eu-publishes-digital-omnibus-leaving-ai-act-future-uncertain