Key Takeaways
- Current and former federal antitrust officials have expressed skepticism that AI companies need a special antitrust exemption to cooperate on safety.
- DOJ’s reported consideration of updated guidance, together with its existing business-review process, offers a possible path for companies seeking greater certainty about defined safety collaborations.
- The 2014 cybersecurity policy statement provides a useful model for technical threat and vulnerability sharing, but it does not necessarily resolve coordination concerning AI development pace, output, capabilities, or release timing.
- The labor-exemption analogy underscores the need to distinguish statutory exemptions from agency guidance, safe harbors, advisory opinions, and enforcement discretion.
- Companies should define the proposed collaboration narrowly, limit competitively sensitive exchanges, preserve independent decision-making, adopt participation and governance safeguards, and assess potential exclusionary effects.
Recent public discussion has intensified debate over whether frontier AI companies need antitrust exemptions, safe harbors, or other regulatory protection to collaborate on safety. Industry participants, policymakers, and commentators have suggested that certain forms of safety-related coordination could benefit from additional legal certainty, including through government-facilitated discussions, agency guidance, safe harbors, or other narrowly tailored mechanisms. At its core, the debate revolves around whether companies need greater certainty, clarity, or protections to coordinate on AI governance, safety, security, and risk mitigation.
The response from current and former federal antitrust officials has been generally skeptical, but has also left room for appropriately structured safety cooperation under existing laws. Current Federal Trade Commission (FTC) Chairman Andrew Ferguson expressed concern: “Regulation plus antitrust exemption sure sounds like moat-digging,” he told the audience at a recent event in Washington, DC.1 Former FTC Chair Lina Khan, meanwhile, argued that existing laws were sufficient to address “dangerous, unvetted or defective products” and that “there’s no AI exemption from laws already on the books.”2
At an event in New York, Associate Attorney General Stanley Woodward, current head of the U.S. Department of Justice (DOJ) Antitrust Division, also suggested that the existing antitrust framework was sufficient to accommodate coordination on safety concerns and invited frontier laboratories to raise concrete concerns with DOJ: “To the extent that anybody at any of the frontier labs wants to come see us about a concern, they would get a meeting,” Woodward said. 3 And former DOJ Antitrust Division head Jonathan Kanter argued that while AI firms “do not need to coordinate in order to deliver safe and secure products to society,” “legitimate collaboration in the context of safety and security” does not require an exemption “because the antitrust laws don’t prohibit that kind of work.”4
The Cybersecurity Information-Sharing Framework Offers a Potential Model
In expressing his receptiveness to engage on AI safety coordination issues, Woodward alluded to the FTC and DOJ’s 2014 Antitrust Policy Statement on Sharing of Cybersecurity Information.5 Against the backdrop of the Cybersecurity Information Sharing Act of 2015,6 the statement explained that properly designed sharing of cyber threat information was unlikely to raise antitrust concerns and could produce substantial security benefits.
The statement focused on sharing technical cyber threat information, such as indicators, threat signatures, and defensive measures, rather than competitively sensitive information concerning price, output, customers, costs, or strategic plans. It therefore offered some degree of comfort for safety-focused collaborations while preserving the broader antitrust enforcement framework.
The cybersecurity framework, combined with current and former antitrust officials’ statements, suggests that regulators believe existing enforcement tools short of an outright exemption may suffice. Updated guidance, a policy statement, advisory opinions, or DOJ business-review letters could clarify permissible AI safety collaboration while allowing enforcement against conduct that restrains competition. Although such guidance can only dictate agency enforcement discretion, it can be useful to courts as persuasive authority in private litigation and state antitrust enforcement.
The FTC’s Labor Exemption Policy Statement Demonstrates the Challenges of Expanding Exemptions
Similarly, the FTC’s January 2025 Enforcement Policy Statement on Exemption of Protected Labor Activity by Workers from Antitrust Liability illustrates the limits of expanding current exemptions to cover new issue areas.7 There, the Commission interpreted an existing exemption in light of statutory text, congressional policy, and the asserted need to protect worker organization.
The labor statement is relevant less for its subject matter than for what it shows about institutional choices. Agencies may use policy statements to explain how they understand an existing exemption or enforcement boundary. But an agency interpretation cannot create a new statutory exemption, and efforts to broaden antitrust carveouts predictably invite disputes about statutory authority, administrability, and effects on nonparticipants.
That experience cautions against treating “exemption,” “safe harbor,” and “guidance” as interchangeable. Each provides a different degree of protection and raises different legal and policy questions. The tension lies in what frontier model companies and government regulators and enforcers may want: While an outright exemption may provide greater clarity and certainty to frontier model companies, enforcers and regulators do not appear eager to grant an exemption that may constrain their discretion and flexibility in a fast-moving industry.
AI Safety Coordination May Present More Complicated Questions Than Cybersecurity Vulnerability Sharing
AI safety collaboration can resemble cybersecurity sharing when it concerns discrete threat indicators, model vulnerabilities, malicious uses, incident reports, or defensive techniques. Those subjects may be important to AI developers, deployers, and critical-infrastructure companies that need to identify and mitigate model-related or cyber vulnerabilities.
Other forms of AI coordination are closer to the competitive line. Discussions about model capabilities, release timing, training investment, deployment conditions, technical standards, and the pace of innovation may implicate output, product design, and strategic decision-making. Within this framework, while each individual company can choose to “pace” its own models to ensure safety and security, a coordinated slowdown among competitors may raise greater concerns.
That distinction explains much of the current skepticism. Ferguson’s concern is that regulation combined with an exemption could facilitate “moat digging.” Kanter’s reported position similarly distinguishes lawful threat-sharing mechanisms from an exemption that might shelter what enforcers view as cartel-like conduct. While certain proposals are framed as narrow and limited to specified safety conversations, a broader proposal to coordinate the speed of frontier development would require enforcers and regulators to evaluate and distinguish safety information sharing from agreements that limit output, product development, or release timing.
Existing Frameworks Can Accommodate Procompetitive Collaborations
U.S. competition enforcers seem to agree, for now, that existing laws and frameworks can accommodate safety collaborations, joint testing, threat-information exchanges, and other integrations that are reasonably necessary to achieve procompetitive or safety-enhancing objectives. But the legality of any arrangement remains fact-specific, and safety rationales do not automatically immunize restrictions on competition. The ultimate question is whether these existing frameworks can accommodate the nature of coordination necessary to ensure safe AI development, or if regulators or enforcers must go beyond statements and guidance. One consideration is that while regulators may respect the bounds of antitrust law that accommodate legitimate safety coordination, aggressive private litigants may still attempt to punish AI firms for taking such safety measures.
Footnotes
1. Jody Godoy, FTC Chair Suspicious of Calls for AI Antitrust Exemptions, Reuters (Sept. 15, 2026), https://www.reuters.com/world/ftc-chair-suspicious-calls-ai-antitrust-exemptions-2026-09-15/.
2. Lina Khan (@linakhanFTC), X (Sept. 14, 2026 at 2:13 ET), https://x.com/linamkhan/status/2099204390548639960.
3. Leah Nylen, US Weighs Antitrust Guidance on AI Safety, Top DOJ Official Says, Bloomberg, republished in Insurance Journal (Sept. 21, 2026).
4. Jonathan Kanter on What the AI Slowdown Means for Competition, The Verge, Decoder Podcast (Sept. 19, 2026), https://www.theverge.com/podcast/997382/openai-microsoft-anthropic-elon-musk-cartel-ai-competition.
5. U.S. Dep’t of Just. & Fed. Trade Comm’n, Antitrust Policy Statement on Sharing of Cybersecurity Information (Apr. 10, 2014), https://www.ftc.gov/system/files/documents/public_statements/297681/140410ftcdojcyberthreatstmt.pdf.
6. See Cybersecurity Information Sharing Act of 2015, Pub. L. No. 114-113, div. N, §§ 101-110, 129 Stat. 2936, 2936-55 (2015) (codified principally at 6 U.S.C. §§ 1501-1510).
7. Fed. Trade Comm’n, Enforcement Policy Statement on Exemption of Protected Labor Activity by Workers from Antitrust Liability (2025), https://www.ftc.gov/system/files/ftc_gov/pdf/p251201laborexemptionpolicystatement.pdf.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
[View Source]
