# OpenAI‘s Threat to Leave Europe Highlights Tensions Over AI Regulation

- Canonical: https://33rdsquare.com/openai-threatens-to-cease-operating-in-europe/
- Published: 2024-09-03
- Author: Jordan Brown
- Categories: [Artificial Intelligence & Machine Learning & ChatGPT](https://33rdsquare.com/category/tech/ai/)

---

The meteoric rise of OpenAI and its viral AI chatbots like ChatGPT and GPT-4 has thrust artificial intelligence into the spotlight. Policymakers are scrambling to put rules in place to govern the rapidly advancing technology, sparking a high-stakes debate over how to balance innovation and safety. The battle lines were starkly drawn in May when OpenAI CEO Sam Altman warned that the company may stop offering its services in the European Union if the bloc‘s proposed AI Act creates an untenable regulatory environment.

"We‘ll try to comply but if we can‘t comply we‘ll cease operating," Altman stated during a visit to London. He argued that some of the technical requirements in the EU legislation would be infeasible for OpenAI to implement. "We think the EU AI Act is a very important initiative that will serve as a key reference point for AI governance globally. But we have some concerns that I don‘t think are fully appreciated about just how hard it is to comply with all of it," he elaborated.[^1]

Altman‘s comments reverberated through the halls of power in Brussels and Silicon Valley. The prospect that arguably the world‘s highest-profile AI company, the creator of models that have enthralled the public and stoked an international AI arms race, could pull out of a market of 450 million consumers raised alarm bells on both sides of the Atlantic. It underscored the immense challenge and high stakes of attempting to regulate a transformative but poorly understood technology that evolves at breakneck speed.

## The EU AI Act Takes Shape

The European Union has emerged as a trailblazer in AI governance, seeking to set global standards as it has for data privacy with its landmark GDPR legislation. The EU‘s proposed AI Act is the world‘s most ambitious attempt yet to regulate AI through a comprehensive risk-based framework.[^2] The act‘s core aim is to protect citizens‘ rights, prevent AI harms, and foster public trust by enforcing strict guidelines around transparency, oversight, and accountability.

Under the AI Act, AI applications would be classified into risk categories:

- _Unacceptable risk systems_, such as those that manipulate human behavior or conduct social scoring, would be banned outright.
- _High-risk applications_, including AI used in education, employment, credit-scoring, and critical infrastructure, would be subject to requirements around data quality, record-keeping, transparency, and human oversight. Providers would need to register these systems in an EU database.
- _Limited and minimal risk systems_, a category covering AI like chatbots and image generators, would face transparency obligations but otherwise remain largely unregulated.[^3]

The European Parliament overwhelmingly voted to adopt the AI Act on June 14, setting the stage for approval by member states later this year. The EU has framed the legislation as essential to mitigating AI risks and harms while providing legal certainty to foster investment and adoption. "This is a significant step towards making trustworthy AI a reality," stated Dragos Tudorache, the MEP co-leading Parliament‘s work on the file.[^4]

## OpenAI Objects

But as the AI Act enters the final stages, major tech companies are sounding the alarm that it could hamper innovation and cement China‘s lead in AI. OpenAI in particular has emerged as a vocal critic, leveraging its clout and popularity to pressure EU policymakers.

In addition to Altman‘s threat to cease EU operations, OpenAI has reportedly lobbied Members of European Parliament and European Commission staff to water down the act‘s provisions.[^5] The company takes issue with requirements around providing detailed technical documentation, conducting conformity assessments, and enabling human oversight for its models, which it argues fail to account for the realities of developing large language models.

OpenAI‘s stance highlights the difficulties of imposing a regulatory framework on general-purpose AI systems like ChatGPT, which can be adapted for a wide range of applications across industries. It‘s unclear whether OpenAI‘s models would fall under the high-risk designation given their potential for both beneficial and harmful uses. The company wants to avoid a scenario where it would need to seek authorization each time a customer deploys its models for a high-risk use case.

At the same time, many have criticized OpenAI‘s lack of transparency around the development of its headline-grabbing models. Unlike some AI labs, OpenAI has not released key details about the training data, architecture, and capabilities of GPT-3 or GPT-4, making it difficult to independently assess their performance and limitations.[^6] This opacity seems to be at odds with the disclosure requirements enshrined in the AI Act.

## A Transatlantic Tug-of-War

OpenAI is not alone in its misgivings about the EU‘s approach. The AI Act has become a lightning rod for trade tensions between the U.S. and Europe, with Washington warning that it could unfairly disadvantage American companies and amount to a discriminatory market access barrier.[^7]

In March, the U.S. Commerce Department sent a strongly-worded letter to the European Commission, arguing that an overly broad definition of AI, cumbersome compliance requirements, and restrictions on non-EU providers "could result in significant market disruptions, stifle innovation, and undermine the objectives of the Artificial Intelligence Act."[^8] Several U.S. agencies and congressional committees have raised similar concerns.

For its part, the EU has pushed back on accusations of protectionism, asserting that the rules would apply equally to European and foreign companies. Officials note that the AI Act does not require data localization and allows for conformity assessments by non-EU bodies. The divergence reflects a broader philosophical split, with the EU taking a more precautionary approach grounded in its commitment to human rights and the U.S. prioritizing innovation and market forces.

Amidst the transatlantic wrangling, Europe has also emerged as a key battleground in the escalating rivalry between the U.S. and China in AI. Chinese companies like Baidu, Alibaba, and Tencent have made significant inroads in European AI research and commercial applications. But the AI Act could constrain their operations, with a provision that would impose additional requirements on high-risk AI systems developed by companies based in authoritarian countries.[^9] The geopolitical stakes of who writes the rules for the global AI economy are only rising.

## Toward a Balanced Approach

With the EU poised to finalize the AI Act by year‘s end, the window is narrowing to find a balanced approach that protects citizens while enabling beneficial innovation. Policymakers face a daunting task given the complexity of the technology and the pace of change. But the stakes of getting AI governance right couldn‘t be higher.

To chart a path forward, governments and companies will need to work together in good faith to resolve key sticking points and flesh out the details of the AI Act‘s requirements. This will require intensive multistakeholder engagement and technical work to develop standards, best practices, and tools for auditing and testing AI systems.[^10] Building public trust will also demand greater transparency and accountability from AI developers like OpenAI.

There are promising efforts underway to foster international cooperation and consistency in AI governance. The OECD‘s AI Principles, adopted by over 50 countries, provide a high-level framework for trustworthy AI grounded in human rights, fairness, transparency, robustness, and accountability.[^11] The Global Partnership on AI, an initiative of the G7, is working to share best practices and align policies across borders. Industry groups like the Partnership on AI are also developing guidance and tools to operationalize ethical AI principles.

But voluntary norms and self-regulation can only go so far in such a high-stakes domain. As AI systems become more powerful and pervasive, hard laws and enforcement mechanisms will be critical to safeguard citizens and earn public confidence.[^12] The EU AI Act, for all its flaws and detractors, is a monumental step in that direction that could set a global precedent. Other governments, from the U.S. to China to Canada and beyond, are watching closely as they develop their own national AI strategies and policies.

The path forward on AI governance is sure to be winding and contested. Striking the right balance between the immense benefits and risks of the technology is one of the great challenges of our time. Sam Altman and his fellow AI pioneers must recognize their profound responsibilities as stewards of increasingly powerful systems. Policymakers must engage in earnest with technical experts to craft smart and agile rules of the road. And the public must be empowered to weigh in on the momentous choices ahead.

If we get it right, the AI revolution could usher in a new era of prosperity, discovery, and flourishing. If we get it wrong, we risk unleashing AI systems that trample rights and sow chaos. The future is not yet written, but it is ours to shape – before it shapes us.

[^1]: Vincent, J. (2023, May 24). OpenAI CEO Says Company May Leave Europe Due to AI Regulations. The Verge. [https://www.theverge.com/2023/5/24/23737878/openai-ceo-sam-altman-leave-europe-ai-regulation](https://www.theverge.com/2023/5/24/23737878/openai-ceo-sam-altman-leave-europe-ai-regulation)
 [^2]: European Commission. (2021, April 21). Proposal for a Regulation Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act). [https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A52021PC0206](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A52021PC0206)
 [^3]: European Parliament. (2022, October 20). Artificial Intelligence Act: Amendments Adopted by the European Parliament. [https://www.europarl.europa.eu/doceo/document/TA-9-2022-0388_EN.html](https://www.europarl.europa.eu/doceo/document/TA-9-2022-0388_EN.html)
 [^4]: European Parliament. (2023, June 14). AI Act: MEPs Adopt Plans to Boost AI and Support Innovation. Press Release. [https://www.europarl.europa.eu/news/en/press-room/20230613IPR80601/ai-act-meps-adopt-plans-to-boost-ai-and-support-innovation](https://www.europarl.europa.eu/news/en/press-room/20230613IPR80601/ai-act-meps-adopt-plans-to-boost-ai-and-support-innovation)
 [^5]: O‘Brien, C. (2023, May 22). OpenAI Is Lobbying Lawmakers to Water Down EU‘s AI Act. VentureBeat. [https://venturebeat.com/business/openai-is-lobbying-lawmakers-to-water-down-eus-ai-act/](https://venturebeat.com/business/openai-is-lobbying-lawmakers-to-water-down-eus-ai-act/)
 [^6]: Heaven, W. D. (2023, June 5). The Inside Story of How ChatGPT Was Built From the People Who Made It. MIT Technology Review. [https://www.technologyreview.com/2023/06/05/1075010/openai-inside-story-chatgpt-built/](https://www.technologyreview.com/2023/06/05/1075010/openai-inside-story-chatgpt-built/)
 [^7]: Espinoza, J. (2023, March 13). US Warns New EU AI Rules Could Stifle Innovation and Weaken Security. Financial Times. [https://www.ft.com/content/0a3f53a5-3ce6-46e7-92bf-3471c1ae33d5](https://www.ft.com/content/0a3f53a5-3ce6-46e7-92bf-3471c1ae33d5)
 [^8]: U.S. Department of Commerce (2023, March 6). Letter to the European Commission on the Artificial Intelligence Act. [https://www.commerce.gov/sites/default/files/2023-03/US-DOC-Letter-on-EU-AI-Act.pdf](https://www.commerce.gov/sites/default/files/2023-03/US-DOC-Letter-on-EU-AI-Act.pdf)
 [^9]: Renda, A., et al. (2022). Artificial Intelligence and Cybersecurity: Opportunities and Challenges for the EU. Centre for European Policy Studies. [https://www.ceps.eu/wp-content/uploads/2022/05/CEPS-TFR_Artificial-Intelligence-and-Cybersecurity.pdf](https://www.ceps.eu/wp-content/uploads/2022/05/CEPS-TFR_Artificial-Intelligence-and-Cybersecurity.pdf)
 [^10]: Fjeld, J., et al. (2020). Principled Artificial Intelligence: Mapping Consensus in Ethical and Rights-based Approaches to Principles for AI. Berkman Klein Center for Internet & Society. [https://cyber.harvard.edu/publication/2020/principled-ai](https://cyber.harvard.edu/publication/2020/principled-ai)
 [^11]: OECD. (2019, May 22). Recommendation of the Council on Artificial Intelligence. OECD Legal Instruments. [https://legalinstruments.oecd.org/en/instruments/OECD-LEGAL-0449](https://legalinstruments.oecd.org/en/instruments/OECD-LEGAL-0449)
 [^12]: Calo, R. (2023). The Case for Hard Law on AI. IEEE Transactions on Technology and Society, 4(1), 2-7. [https://ieeexplore.ieee.org/document/10057204](https://ieeexplore.ieee.org/document/10057204)

---

Source: [OpenAI‘s Threat to Leave Europe Highlights Tensions Over AI Regulation](https://33rdsquare.com/openai-threatens-to-cease-operating-in-europe/)