Proposal to slow AI development sparks debate over the limits of cooperation among competitors and the role of antitrust authorities
On September 12, 2026, Dario Amodei, CEO of Anthropic, publicly advocated for a coordinated slowdown in the development of the most advanced Artificial Intelligence (AI) models. In an open letter, he argued that competition among companies could drive AI systems to advance faster than the testing and safety mechanisms needed to mitigate their risks. The proposal quickly gained support from executives at competing companies, including OpenAI, SpaceXAI, and Google DeepMind.
As expected, it did not take long for Donald Trump to intervene in the debate. On September 29, the White House brought together industry leaders and signed the White House Accord on Superintelligence, a commitment to establishing safety mechanisms, internal controls, and independent external audits for AI. The document states that "participating companies will meet regularly to establish standards and best practices to enhance the safety of their systems" (free translation), without specifying whether such collaboration would involve slowing down the pace of AI model development.
Meanwhile, the controversy also reached the U.S. courts. Four consumers filed a class-action lawsuit against the companies mentioned above, alleging that their support for Amodei's letter suggested a coordinated effort to delay advances in AI models, potentially violating U.S. antitrust law.
The lawsuit acknowledges the risks associated with AI but argues that each company should act independently by expanding safety testing and engaging external evaluators, rather than agreeing to a coordinated slowdown. According to the plaintiffs, such an agreement would harm consumers and produce effects similar to those of a cartel. They further argue that, if necessary, any coordinated slowdown should be determined by lawmakers and regulatory agencies, rather than by the companies themselves.
Recently, researchers and former employees of AI companies have raised concerns about the risks that rapid AI advancement could pose to humanity's very survival, including the loss of control over information systems, automated warfare, and the development of biological weapons, among others.
This scenario raises important questions: To what extent can competitors cooperate to ensure AI safety without undermining competition and innovation? Who should define these boundaries in the absence of specific regulations?
In Brazil, the antitrust authority, the Administrative Council for Economic Defense (CADE), has recently signaled that it is closely monitoring developments in the AI sector, particularly by reviewing talent acquisitions (acqui-hires) and business partnerships that did not meet the mandatory notification requirements under the Brazilian Competition Law (Law No. 12,529/2011). An investigation is also underway against Meta concerning alleged restrictions on access to AI tools, alongside inquiries into algorithmic collusion. However, to date, the pace of AI model development has not been the subject of a specific CADE investigation.
Article 36, paragraph 3, item VIII, of the Brazilian Competition Law provides that agreements to "limit or control research and technological development" may constitute anticompetitive conduct. Therefore, any proposal to coordinate the pace of AI innovation raises competition concerns and requires an assessment of the permissible boundaries of such agreements.
These concerns have already been reflected in CADE's approach. In a case involving Amazon and Anthropic, CADE's Tribunal emphasized that collaboration agreements in the AI sector may be investigated for potential antitrust violations, even when they do not qualify as transactions subject to mandatory merger notification.
CADE's experience in other economic sectors may offer valuable insights. The authority has demonstrated openness to collaborative initiatives, provided that appropriate parameters and safeguards are established, including restrictions on the exchange of commercially sensitive information.
During the pandemic, for example, CADE approved Movimento Nós, a temporary collaboration among companies such as Ambev, Coca-Cola, and Nestlé aimed at supporting small retailers in safely reopening their businesses. In its assessment, CADE considered the initiative's specific purpose, duration, and contractual limitations, ensuring that it did not involve commercial coordination or the exchange of competitively sensitive information.
This precedent demonstrates that exceptional circumstances may lead antitrust authorities to permit cooperation among competitors. In the context of AI, the identified risks could justify joint safety measures, provided that such initiatives are limited to what is necessary and preserve each company's autonomy to innovate and develop its own models.
To provide greater predictability in the assessment of such agreements, CADE launched a public consultation on its Guidelines for Cooperation Among Competitors in March 2026, which concluded on June 18. The contributions received remain under review by the authority.
Brazil does not yet have specific AI legislation to guide risk assessments. Nevertheless, competition authorities have the power to assess the legality of agreements among competitors aimed at establishing common AI safety standards.
The urgency of this issue reinforces the need to carefully assess cooperation among companies and makes it essential that any jointly agreed limitations on technological innovation be subject to regulatory scrutiny, rather than being determined solely by competitors.
Marcelo Proença, a partner at Manesco Advogados, holds a Ph.D. and a Master's degree in Commercial Law from the University of São Paulo Law School (USP). He is a professor at USP and FGV Law, where he teaches courses in Compliance, Corporate Law, and Competition Law.
Lucas de Góis Barrios, a senior associate at Manesco Advogados in Brasília, is a Ph.D. candidate in Commercial Law at USP. He holds a postgraduate specialization in Economic Law from FGV-SP and a Bachelor of Laws degree from the Federal University of Bahia (UFBA), including a period of study at Sciences Po Paris. He has extensive experience in competition and regulatory matters, having held strategic positions in the public sector, where he led and advised on complex cases, major institutional projects, administrative enforcement proceedings, and settlement negotiations.
Pedro Henrique Mariz de Oliveira, a intern at Manesco Advogados and a law student at the University of Brasília.
