Cohere CEO Warns Against an AI Safety ‘Cartel’

Cohere CEO Aidan Gomez warned that allowing the largest frontier-AI labs to coordinate safety standards could entrench Anthropic, OpenAI and other incumbents in a potential “cartel.” Gomez supports tougher safeguards, independent testing and transparency, but argues that AI rule-setting should involve a broader stakeholder group. The comments underscore rising competitive and regulatory tensions around governance of frontier AI.

Analysis

The investable issue is not near-term AI safety spending; it is whether regulatory compliance becomes a fixed-cost moat. A regime designed by incumbent frontier labs would favor vertically integrated players with proprietary evaluation infrastructure, legal teams, privileged cloud capacity, and government relationships. That outcome would reinforce the relative positioning of MSFT/OpenAI, GOOGL, AMZN/Anthropic and META, while raising the capital required for application-model challengers and independent model developers.

The more plausible 1-3 month catalyst is regulatory process rather than a binding rule: consultation language, procurement standards, or evaluation requirements can signal which firms' existing governance stacks become de facto benchmarks. Broad stakeholder participation would reduce the probability of an explicit incumbent cartel, but could also produce fragmented standards that slow enterprise deployments; this is modestly negative for near-term AI monetization multiples at software names whose valuations assume rapid adoption. Watch EU AI Act implementation guidance and U.S. federal-agency procurement language rather than executive commentary.

Contrarian view: market participants may overestimate the benefit to frontier-model vendors from tighter rules. Enterprise buyers already demand auditability, indemnification, data controls and model-risk governance, so mandatory standards could commoditize trust features rather than create durable pricing power. The larger structural beneficiary over 6-18 months may be cybersecurity, testing and governance vendors—PANW, CRWD, PLTR and ServiceNow (NOW)—if compliance shifts from bespoke lab processes into recurring enterprise workflow spend.

There is no standalone trade from this commentary. The thesis is falsified if regulatory proposals explicitly exempt open-weight models or impose uniform, low-cost self-attestation; that would weaken the compliance-moat narrative and favor lower-cost/open ecosystems over closed-model incumbents.

Key Decisions for Investors

• Maintain a watchlist rather than initiate a directional AI-model trade; require a concrete U.S. procurement rule, EU implementation standard, or formal industry code with testing obligations before positioning.
• On evidence that third-party testing, audit trails or ongoing monitoring become mandatory, accumulate PANW and NOW over 1-3 months versus a basket of high-multiple application software; recurring compliance budgets should be more durable than discretionary AI pilots.
• Use MSFT versus a broad software proxy such as IGV as a conditional pair trade only if rules privilege certified closed-model providers: target a 3-6 month holding period, with exit on language endorsing open-model self-certification or on material Azure AI growth deceleration.
• Monitor enterprise AI contract disclosures for indemnification and governance attach rates. If these features do not translate into higher ACV or services revenue by the next two earnings cycles, avoid assigning a regulatory premium to AI infrastructure or governance vendors.