Behind closed doors in Washington, artificial intelligence rivals that usually battle for talent and benchmark supremacy have spent weeks meeting in secret. Can fierce commercial adversaries set technical boundaries before catastrophic computational risks materialize? OpenAI policy chief Chris Lehane confirmed on Tuesday that OpenAI has joined competitors Anthropic and Google DeepMind in ongoing AI safety coordination to establish shared defensive benchmarks. The disclosure comes after Anthropic CEO Dario Amodei urged the tech sector to curb frontier development speeds, warning that unchecked momentum invites disaster. The private alliance has triggered immediate regulatory resistance from federal officials who warn that collaborative industry slowdowns could violate federal antitrust statutes. [1, 2]
Rivals Unite for Unprecedented Safety Discussions
Chris Lehane delivered his remarks during a series of Capitol Hill briefings with American lawmakers, confirming reporting first aired by Bloomberg. The discussions reflect mounting alarm. Generative capabilities advance without clear technical boundaries. Anthropic CEO Dario Amodei sparked the public debate on Saturday by releasing an extensive manifesto that pleaded for collective restraint. OpenAI chief executive Sam Altman, Google DeepMind leader Demis Hassabis, and SpaceXAI founder Elon Musk quickly backed Amodei’s call to action. In an interview with Fortune late last week, Altman hinted at confidential discussions occurring across the frontier sector. [1, 2]
Cooperation among commercial rivals marks a striking shift for an industry defined by relentless speed. As part of this emerging framework, Altman announced that OpenAI will join Anthropic in embedding independent third-party evaluators directly within corporate infrastructure to oversee model safety. A subsequent report from The Information revealed that OpenAI, Anthropic, and Google DeepMind intend to form a permanent standards body for the entire ecosystem. Altman reportedly informed employees that tech leaders must build this supervisory mechanism independently, operating without immediate federal assistance. Demis Hassabis had previously urged United States authorities in July to construct an oversight body capable of evaluating frontier models and orchestrating uniform industry pauses. [1, 2]
Fierce commercial competition makes such joint technical commitments complicated to maintain in practice. PerEXP Teamworks previously analyzed proposals to slow the AI race through outside oversight, which emphasized how hard unmonitored agreements are to police. Now, ongoing AI safety coordination forces engineering teams to share sensitive protocols while racing for product market share. Can historic competitors sustain operational trust? That tension defines the new alliance. [1, 2]
Antitrust Law Challenges AI Safety Coordination
Antitrust scrutiny arrived quickly. Legal scholars and competition attorneys warn that joint agreements to slow model development could trigger severe penalties under the Sherman Act, the bedrock federal antitrust statute. When leading competitors coordinate technical roadmaps or delay commercial releases, enforcers can classify their actions as output restriction (agreements where competitors restrict technical pacing or capacity). Even benevolent safety goals do not automatically shield corporate cartels from federal antitrust enforcement. Altman himself admitted during recent industry discussions that informal technical pacts risk crossing into illegal market collusion. [1, 2]
Dario Amodei recognized this legal obstacle in his published essay. He directly requested that the United States government intervene to “mediate or at least enable” multilateral cooperation, suggesting that federal agencies issue formal legal waivers to protect companies from antitrust prosecution. Without official protection, corporate attorneys might block engineers from participating in joint technical panels. Statutory competition laws leave little room for private compacts. [1]
Federal antitrust authorities remain highly skeptical of private corporate pacts. Regulatory officials caution that voluntary pledges between dominant market leaders rarely protect consumers. [1]
Debating Waivers and Historic Industry Precedents
Chris Lehane offered a sharply contrasting legal perspective during his Washington press conference, telling Bloomberg that OpenAI does not anticipate needing a government antitrust waiver to continue AI safety coordination. The policy chief argued that existing statutory frameworks already provide sufficient leeway for competitors to collaborate on urgent public safety issues. Lehane cited commercial aviation as a primary precedent, noting that rival airline carriers routinely share proprietary flight and equipment data to eliminate catastrophic navigational hazards without facing antitrust penalties. [1]
Historical precedent extends beyond commercial aviation into federal cybersecurity policy. Legal scholars point out that the Obama administration established specific guidance permitting technology corporations to exchange threat intelligence to defend critical digital infrastructure. Proponents argue that managing frontier model risks falls squarely under similar national defense doctrines, rendering special legislative exemptions unnecessary. The strategic decision by OpenAI to prioritize safety also aligns with internal corporate shifts; PerEXP Teamworks previously covered putting OpenAI’s potential IPO on hold to prioritize safety amid rapid technical transformation. Corporate priorities clash with external legal requirements. [1]
Senior enforcement officials in the Trump administration reject the airline analogy. Federal Trade Commission Chairman Andrew Ferguson expressed deep skepticism during an address at Georgetown University, as reported by Reuters. Ferguson stated that all of my alarm bells go off whenever corporations arrive in the capital simultaneously requesting federal regulation and antitrust exemptions. Dominant firms use these demands to build high barriers to entry that insulate incumbents from competitive challenge. Attorney General Todd Blanche told Bloomberg that the Justice Department will consult the wider administration before determining whether any antitrust exemption serves presidential priorities. [1]
Congressional Bipartisan Bills Face White House Skepticism
Lawmakers in Congress have introduced competing legislative solutions to address model oversight and antitrust risk. In July, Indiana Senator Jim Banks and California Senator Adam Schiff introduced the Collaboration on Adversarial Threats and Security Risks Act, a bill offering explicit antitrust carveouts (statutory exemptions shielding collaborative conduct from competition penalties) to enable private AI safety coordination against catastrophic hazards. The proposal aims to establish safe harbors where corporate researchers can share safety diagnostics without risking federal prosecution. [1]
OpenAI is actively lobbying for alternative legislative frameworks on Capitol Hill. Lehane confirmed that the company backs the bipartisan FRONTIER Act, which would establish a formal federal safety structure and mandate that top frontier labs submit to audits by independent verification organizations (non-governmental entities authorized to audit frontier AI models). However, the measure faces stiff resistance from members of the House AI Commission. A second proposal under development by Senate Majority Leader John Thune, Texas Republican Senator Ted Cruz, and Minnesota Democratic Senator Amy Klobuchar would empower the Commerce Secretary to dispatch government auditors to test leading frontier models. [1]
Time is running short. Legislative passage appears increasingly improbable before the congressional session concludes. Senate Majority Leader John Thune told Politico that passing comprehensive technology legislation this calendar year presents severe logistical challenges given congested legislative dockets. Thune stated that while conversations regarding potential paths forward will continue, accomplishing major statutory overhauls in the near term remains difficult. President Donald Trump remains firmly opposed to imposing regulatory constraints on artificial intelligence laboratories, insisting that federal mandates undermine American commercial momentum. [1, 2]
Fierce Startup Pushback and Geopolitical Rivalry
Independent artificial intelligence startups view these regulatory overtures as a coordinated anti-competitive maneuver. Cohere chief executive Aidan Gomez published an essay attacking the safety coalition, characterizing calls for antitrust waivers as an effort by dominant incumbents to monopolize regulatory standards. Gomez argued that oligopolies exploit public fear to bend statutory competition rules and dictate terms for smaller competitors, describing the coalition as a wolf in sheep’s clothing, a cartel by another name. Emerging competitors argue that voluntary AI safety coordination among incumbent giants would effectively lock out challengers lacking vast capital resources. [1]
White House leadership shares this skepticism, framing domestic regulatory constraints as an unacceptable geopolitical risk. President Trump has publicly dismissed catastrophic risk warnings as an exaggerated hoax, asserting that any mandatory slowdown in development speeds would cede technological leadership to China. Key White House technology advisor David Sacks, an active venture investor holding substantial financial stakes across the sector, has reinforced this position by arguing that existential risk arguments are grossly overblown. Federal policy accelerates commercial innovation rather than codifying technical restrictions. [1, 2]
The collision between safety pledges, antitrust statutes, and geopolitical pressure leaves tech executives navigating uncharted legal territory. As OpenAI, Anthropic, and Google DeepMind advance their private alliance, federal regulators show no appetite for granting informal antitrust carveouts. Industry leaders must now decide whether to formalize their joint technical benchmarks without federal protection or abandon collaborative oversight under the threat of aggressive antitrust prosecution. The outcome remains uncertain. For the immediate future, voluntary AI safety coordination must survive in the shadow of the Sherman Act. [1, 2]
- ONLINE NEWS Yildirim, E. (2026, September 15). OpenAI, Anthropic, and Google Join Forces for Safety as Antitrust Concerns Mount. Gizmodo. [Article Link]
- ONLINE NEWS Bellan, R. (2026, September 15). OpenAI, Anthropic, Google have been in talks on AI safety for weeks. TechCrunch. [Article Link]
2 comments