CEI Comments on Collaboration Guidelines Request for Information

The Competitive Enterprise Institute (CEI) appreciates the opportunity to comment on the Federal Trade Commission (FTC) and the Department of Justice (DOJ) Antitrust Division’s Joint Public Inquiry for Consideration of Guidance on Collaborations Among Competitors.  CEI is a nonprofit research and advocacy organization that focuses on regulatory policy from a free-market perspective.

The FTC and DOJ have taken a proactive and welcome step by launching their joint public inquiry regarding potential updated guidance on joint ventures and collaborations among competitors. As both agencies know well, the prior guidelines, the 2000 Antitrust Guidelines for Collaborations Among Competitors (Collaboration Guidelines),  were withdrawn in the final weeks of the previous administration “without providing any replacement guidance, or even intimating plans for future replacement.”

Although there are many areas where the FTC and DOJ should provide helpful guidance for business collaboration, CEI is writing to specifically advocate for guidance on AI safety and security collaboration. The antitrust agencies have a tremendous opportunity to provide clarity for the quickly developing AI sector. To effectively meet this moment, the updated Collaboration Guidelines should establish clear antitrust safe harbors for collaborative efforts around AI safety and cybersecurity. By enabling proactive, industry-led mitigation and information sharing among competitors, the FTC and DOJ can help facilitate the safety of Americans and the country’s technology infrastructure.

AI Joint Ventures and Collaborations

Regarding the FTC and DOJ’s inquiries as to what technologies would benefit and what technological changes should be considered in updates to the Collaboration Guidelines, there is no more notable example than artificial intelligence (AI). In 2025, private AI investment in the U.S. reached $285.9 billion, according to Stanford University’s 2026 AI Index Report.  Globally, AI accounted for 61 percent of all venture capital funding in 2025.  Over the next five years, capital investment in chips, data centers, and power delivery is estimated to reach $4 to $8 trillion.

The FTC and DOJ’s joint statement withdrawing the prior Guidelines noted that the former “Collaboration Guidelines fail[ed] to address the competitive implications of modern business combinations and rapidly changing technologies such as artificial intelligence . . . .” This rationale mirrors the impetus behind the Guidelines’ original drafting. The 2000 Collaboration Guidelines were born out of several years of study and public input, including the FTC’s 1995 Global and Innovation-Based Competition Hearings, which explored whether rapid technological change warranted adjustments in the way the agency enforces antitrust and consumer protection laws.

Some U.S. tech companies have expressed the need for antitrust guidance on collaborations involving AI,  and former U.S. antitrust enforcers have done the same. According to Alden Abbott, former FTC General Counsel, “[m]odern markets—especially those shaped by artificial intelligence—need clear rules that distinguish genuinely harmful collusion from productive, welfare-enhancing cooperation.”

Abbott argues further that research and development (R&D) provides the strongest case for new antitrust safe harbors:

Research joint ventures can internalize spillovers, spread risk, reduce duplicative fixed costs, and combine specialized know-how that no single firm possesses in sufficient measure. Those benefits grow in AI and other nascent technologies, where research is expensive, failure rates are high, and the social value of success can far exceed what any one participant can capture.

This echoes expert testimony from the FTC’s Joint Venture Project hearings, which informed the prior Collaboration Guidelines. In 1997, the FTC hosted five hearings from June 2 to July 1, discussing the unclear application of antitrust law to joint ventures and whether the current state of the law posed a serious impediment to pro-competitive collaborations.  The late Harvey Goldschmid, a professor of law at the Columbia Law School, said during a hearing held on June 2, 1997, that “[t]here is no doubt in my mind and most other minds that joint ventures can enhance efficiency in areas like research and development, production, marketing, and joint purchasing.”  Further, joint ventures and collaborations “allow for economies of scale to be obtained, provide synergies, spread risks, and allow firms to exploit, in a healthy way, each other’s expertise and technological capabilities . . . .”

Courts have long recognized that legitimate joint ventures and research collaborations may generate substantial procompetitive efficiencies and are therefore commonly evaluated under the rule of reason rather than treated as per se unlawful restraints. As the rapid technological shifts of the 1990s prompted the creation of the original 2000 Collaboration Guidelines, the proliferation of AI creates the need to clarify and protect pro-competitive joint ventures and collaborations.

Safe Harbor for Collaboration on AI Safety and Security

The FTC and DOJ should include an antitrust safe harbor for AI safety and security collaborations in their updated Guidelines to foster industry-led risk mitigation and limit the need for heavy-handed regulation. “A safe haven is cost-effective for both industry and the government” and has the “potential to stimulate research without imposing burdensome regulations on industry,” according to Ella Duus, now a fellow at the RAND Center on AI, Security, and Technology.  This safe harbor should apply broadly to both cybersecurity as well as more dynamic AI safety issues like model drift and hallucinations.

Still, as Abbott writes in his comments to this joint inquiry, antitrust safe harbors for collaborations should “distinguish[] between cartel behavior and productive cooperation.”  The agencies can narrowly tailor such safe harbors to preserve robust antitrust enforcement. As Abbott explains,

The guidance should also identify the features that would take a collaboration outside those safe harbors. These include naked agreements on price, output, wages, or customer allocation; restraints unrelated to the venture’s legitimate objectives; exclusionary access rules lacking technical or investment-based justification; and the exchange of competitively sensitive information unnecessary to the collaboration. Framing the analysis this way would give businesses clearer direction while preserving ample room for enforcement against genuinely anticompetitive arrangements.

The announcement of Anthropic’s Mythos model has elevated the conversation around AI and cybersecurity.  The company’s own testing revealed that the model can identify and exploit zero-day vulnerabilities across every major operating system and web browser, having found some vulnerabilities that were decades old.  While some view the reaction to Mythos as hysteria, experts have long recognized that AI will fundamentally transform cybersecurity.  Anthropic launched Project Glasswing as part of its announcement, sharing a preview of their model with companies like Amazon Web Services, Apple, Google, Microsoft, and others to allow for defensive security work.  However, as some antitrust experts point out, “the coalition risks violating Section 1 of the Sherman Antitrust Act.”

Recently, Michael J. Kratsios, Director of the Office of Science and Technology Policy, published a memo stating that “the United States Government has information indicating that foreign entities, principally in China, are engaged in deliberate, industrial-scale campaigns to distill U.S. frontier AI systems,” which includes “[l]everaging tens of thousands of proxy accounts to evade detection and using jailbreaking techniques to expose proprietary information,”  In turn, Kratsios states the Trump Administration will “[e]nable the private sector to better coordinate against such attacks.”  As part of that effort, the antitrust agencies should include an antitrust safe harbor for cybersecurity coordination in their updated Collaboration Guidelines, with appropriate guardrails.

In considering an update to the agencies’ Collaboration Guidelines, the FTC and DOJ should also consider updating their 2014 Antitrust Policy Statement on Sharing of Cybersecurity Information. Despite the decision to withdraw the prior Collaboration Guidelines in December 2024 with no plans to contemplate new ones, the FTC and DOJ prudently noted that the withdrawal did not affect the antitrust agencies’ 2014 Antitrust Policy Statement. However, in light of the proliferation of AI, necessary updates are needed to provide further clarity on the sharing of cybersecurity information.

American tech companies launched the non-profit Frontier Model Forum (FMF) in July 2023 in order to ensure safe and responsible development of frontier AI models. Recently, the Frontier Model Forum published an issue brief entitled Information Sharing, Incident Reporting, and Incident Response for Frontier AI Risks. The issue brief highlights how the 2014 Policy Statement is insufficient in providing clarity for AI developers.

The 2014 Policy Statement focuses on technical cybersecurity information like threat signatures and IP addresses but does not adequately cover more dynamic information addressed in the FMF’s issue brief. The FMF identifies “[v]ulnerabilities, weaknesses, and exploitable flaws” as a primary category for information sharing. It describes such information as “[v]ulnerabilities, weaknesses or exploitable flaws [that] may compromise the safety, security, or intended functionality of frontier AI models. Examples may include jailbreaks, adversarial inputs, data poisoning, or other attempts to bypass model safeguards.” 

In addition to providing a safe harbor for AI security collaboration, the FTC and DOJ should do the same for AI safety collaboration. “AI safety asks ‘will this system behave as intended?’ while AI security asks ‘can someone compromise this system?’ Both matter, but they require fundamentally different approaches,” according to Morne Wiggins at Agility at Scale.  However, AI safety and security may blend when data poisoning or jailbreaking contributes to unsafe model drift or hallucinations. The overlap between AI safety and AI security should be studied by the antitrust agencies in considering updated Collaboration Guidelines.

On the issue of AI safety collaboration, Nicholas Felstead, AI policy fellow at the Center for Law & AI Risk, writes that “AI labs face the fear that cooperation with competitors will trigger antitrust scrutiny from the Department of Justice or Federal Trade Commission (FTC).” And industry members have asked for further guidance on AI safety collaboration. In response to the National Telecommunications and Information Administration’s (NTIA) “AI Accountability Policy Request for Comment,” Anthropic suggested:

Regulators should issue guidance on permissible AI industry safety coordination given current antitrust laws. Clarifying how private companies can work together in the public interest without violating antitrust laws would mitigate legal uncertainty and advance shared goals.

The FTC and DOJ should consider coordinating with cybersecurity and technical agencies when designing these AI collaboration safe harbors to enable voluntary commitments to safety and security.  The Cybersecurity and Infrastructure Security Agency’s (CISA) Joint Cyber Defense Collaborative published its JCDC AI Cybersecurity Collaboration Playbook in January 2025 to help inform operational collaboration and information exchange.  Further, in March 2025, the National Institute of Standards and Technology (NIST) published its report, Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations, which “provides a taxonomy of concepts and defines terminology in the field of adversarial machine learning (AML).”

Conclusion

The FTC and DOJ deserve credit for beginning the process of developing updated Antitrust Guidelines for Collaborations Among Competitors. In light of recent events involving AI and cybersecurity, the private sector should be empowered to coordinate and adapt to a fast-changing digital landscape without the fear of running afoul of antitrust laws. By aligning new antitrust safe harbors with technical benchmarks like those from CISA and NIST, the antitrust agencies will foster important safety and security collaboration while still being able to fulfill their competition missions.

Respectfully submitted,

Alex R. Reinauer
Research Fellow
Competitive Enterprise Institute
[email protected]