The Challenge to Digital Sovereignty: Where is Europe Heading with the Delay in the Implementation of the AI Act?

Prologue: The "Time Bomb" of August 2

This summer, Brussels is abuzz with the metaphor of a "time bomb." August 2 marks the day when the countdown begins for the transparency obligations and penalties applicable to general-purpose AI (GPAI) models under the EU's first comprehensive AI regulation, the "AI Act." However, just before this date, the European Commission admitted to delaying the publication of the "AI Code of Practice" guidelines for businesses by more than six months and is reportedly considering a two-year freeze on the enforcement of the law. In response, the hashtag #StopTheClockAIAct has surged on X (formerly Twitter), sparking a three-way debate among the tech industry, politicians, and civic groups.reuters.com

Chapter 1: The Complexity of the AI Act and "Phased Implementation"

The AI Act was published in the official journal in June 2024 and is already in effect, but penalties and detailed obligations will be applied gradually over a maximum of three years [Figure 1]. Among these, the transparency obligations for general-purpose AI models will take effect on August 2, 2025, with additional regulations for the "systemic risk" category set for August 2, 2026. This "two-stage rocket" creates a moving target for companies, making it difficult for practitioners to plan compliance. The operational standards (EN standardization documents) and the Code of Practice, which the European Commission promised to supplement, have been delayed, resulting in a "document-first" risk where only the legal text is ahead.reuters.com

Chapter 2: The Shock of the 45 Companies' Open Letter

On July 3, 45 major European companies, including ASML, Airbus, SAP, Mercedes, and BNP Paribas, demanded a "two-year clock stop" in an open letter. "Ambiguous and overlapping regulations undermine the competitiveness of the European AI industry," the letter, spanning 35 paragraphs, warns. The letter was supported by the European AI Champions Initiative (AICI), which has over 110 member companies. An article in SiliconANGLE on the same day reported that "hasty regulations will decisively widen the gap with China and the United States," and the Financial Times also featured it on its front page.siliconangle.com

Chapter 3: Political Temperature Differences: The Swedish Prime Minister and Digital Minister

The Nordic countries responded to the cries of companies. Swedish Prime Minister Ulf Kristersson stated that "the provisions are too complex, causing confusion on the ground," supporting the postponement argument. German Digital Minister Volker Wissing also remarked, "If we proceed without guidelines and standards, accidents are inevitable." Furthermore, some members of the European Parliament's Industry Committee are considering including a "simplification package" in an "omnibus amendment" by the end of the year.cio.comlinkedin.com

Chapter 4: The Pressure from the US IT Lobby and CCIA

Major US companies like Alphabet, Meta, and Apple are also aligning with European companies. The IT lobby group CCIA Europe submitted a letter at the June 25 summit, urging to "halt enforcement until standards are in place." An AWS survey cited that "two-thirds of European companies do not understand the obligations." US companies have started to publicly state that they would rather prioritize projects in the US and Asia, where regulations are clearer, than allocate investment resources to the opaque European market.reuters.comcio.com

Chapter 5: The Cry of Startups—Mistral and Black Forest Labs

For small AI startups, this is a matter of survival. French Mistral AI and German Black Forest Labs, among others, jointly signed a statement arguing that "it's unfair for emerging companies with limited funds and manpower to be forced to do the same paperwork as large corporations." In an interview with Sifted, Mirakl's VP lamented, "If the vaguely defined 'substantial modification' criteria differ across countries, we will have to deal with 27 different interpretations."sifted.eu

Chapter 6: Social Media is Divided

The hashtag #StopTheClockAIAct recorded 75,000 tweets on X from July 3 to 5 (Brandwatch estimate). The most retweeted post from supporters stated, "If we reach August 2 with the current legal text, Europe will become an innovation desert" (from @AI_Champs representative). On the other hand, the human rights NGO @Rights4EU countered, "Don't succumb to the lobbying of surveillance capitalism," garnering about 18,000 likes. On LinkedIn, a post by MLex journalist Luca Bertuzzi gathered over 800 reactions, raising the legal question, "Can the Commission unilaterally delay enforcement?"linkedin.com

Chapter 7: Concerns of Consumer Groups and Academic Societies

The European Consumer Organization (BEUC) criticized that "a delay would postpone consumer protection by two years." The European AI Ethics Society (EAIE) warned in a statement that "if transparency reporting is postponed, ordinary citizens will continue to suffer from black-box AI." Scholars propose separating discussions on "technical safety" and "regulatory timeliness," suggesting a compromise that distinguishes between the priority of "risk assessment and audit systems" and "model parameter disclosure obligations."

Chapter 8: The Background of Technical Standard Delays

The AI Act relies on links to over 200 technical standards formulated by the European Committee for Standardization (CEN and CENELEC), but currently, only just over 30% of the drafts have been published. There is insufficient academic consensus on evaluation metrics like "toxicity tests" and "energy efficiency measurements." Commission staff admit that "even if we double the pace of EN standard development, it won't be completed until early 2026." The delay in guidelines merely reflects this reality.

Chapter 9: International Comparison: The "Loose and Fast" of the US and China, the "Cautious and Slow" of the EU

The US promotes private sector-led initiatives through "soft law" combining executive orders and the NIST framework, while China mandates "pre-shipment security reviews" under its interim regulations for generative AI, yet encourages early development with a "regulatory sandbox." In contrast, the EU aims for a unified approach with highly binding regulations, but the cost and time for standard development are bottlenecks. According to estimates by the IFO Institute for Economic Research, the AI compliance burden for EU companies is 1.7 times that of the US and 2.4 times that of China.

Chapter 10: Expert Perspectives: The Dilemma of Risk-Based Supervision

Among regulatory practitioners, the focus is on how to balance "risk-based supervision" with a "hybrid self-declaration model." In an interview with CIO magazine, CCIA Europe representative Daniel Friedlander criticized that "the current provisions impose excessive obligations on model providers and obscure the responsibilities of user companies."

Meanwhile, in an IAPP panel discussion, Irish MEP Michael McNamara stated, "A delay is realistic, but it must be accompanied by enforceable interim guidelines to compensate for the absence of standards, or it will be toothless."cio.comiapp.org

Chapter 11: Scenario Analysis: Three Futures

1. Full Delay Scenario (45% Probability)
   A two-year clock stop plus the establishment of a simplified legal text package. Companies transition to voluntary transparency reporting while waiting for standard development.

2. Limited Delay Scenario (40%)
   Only the GPAI portion is delayed by six months to a year. Other high-risk applications proceed as scheduled. The EU refers to this as a "split launch."

3. Continuation of Current Schedule (15%)
   Enforcement begins on August 2. Guidelines are published by the end of the year. Violation determinations concentrate after August 2026, remaining "symbolic regulation" for the initial three years.

Chapter 12: Practical Recommendations for Companies

• Risk Inventory: Conduct an inventory of GPAI model usage, categorized as "provider/deployer."

• Transparency Template: Pre-publish summaries of training data and energy consumption measurement processes to reduce revision costs when guidelines are finalized.

• Establish Contact Points: Provisionally design incident reporting channels to the European AI Board and national supervisory authorities.

• Participate in Standardization Forums: Incorporate proprietary technology into standard proposals to reduce future compliance testing costs.

Conclusion: The "Long Summer" of Digital Sovereignty

The "human-centered, trustworthy AI" that the EU aims for with the AI Act is a dual goal that is difficult to achieve alongside industrial competitiveness. The battle over delays serves as a "litmus test" highlighting this conflict. When August 2 arrives, will it be the clock or the rules that stop? Only the summer sky over Brussels knows the answer. However, companies have no choice but to advance "voluntary transparency" in preparation for the inevitable enforcement.