US AI regulators vs. frontier labs: Anthropic pushes for model “kill switches” as Visa and ChatGPT move to payments

Anthropic CEO Dario Amodei said the US government should have the authority and capability to block developers from deploying new AI models when they pose specific risks. The comments, reported by kommersant.ru on 2026-06-11, frame safety as a governance problem that requires enforceable state power rather than voluntary restraint. In parallel, Visa is embedding its payment network into ChatGPT so the chatbot can shop and complete transactions for users, turning conversational AI into a direct commerce interface. Separate reporting also claims that some ChatGPT-linked accounts tried to “exploit and amplify existing public concerns” about energy prices, suggesting coordinated or opportunistic information operations around a sensitive macro topic. Geopolitically, the cluster points to a fast-moving contest over who controls frontier AI deployment: governments seeking regulatory “gates” versus companies and platforms expanding capabilities at commercial speed. If US authorities can effectively pause or block model releases, it would reshape bargaining power across the AI supply chain, affecting cloud providers, model developers, and downstream integrators like payment networks. The Visa-ChatGPT payments integration raises the stakes for financial oversight, fraud controls, and cross-border compliance, because AI-driven transactions can scale faster than traditional merchant onboarding and dispute mechanisms. Meanwhile, the alleged energy-price narrative exploitation highlights how AI systems and their user ecosystems can become vectors for social and market destabilization, especially when inflation and utility costs are already politically charged. Market and economic implications are likely to concentrate in fintech rails, digital payments, and AI infrastructure. Visa’s move could increase transaction velocity and expand addressable commerce volume for conversational interfaces, pressuring incumbents in payment orchestration and merchant tooling, while also increasing demand for risk engines and identity verification. The energy-price concern amplification claim signals potential volatility spillovers into utilities, retail energy, and broader inflation expectations, even if the direct economic effect is indirect. For investors, the immediate read-through is that AI governance and trust will become a tradable variable: companies tied to compliance, monitoring, and secure deployment may see a relative bid, while those exposed to reputational or regulatory risk may face a discount. What to watch next is whether US regulators operationalize Amodei’s “block” concept into concrete licensing, evaluation, or deployment controls with measurable thresholds. Track announcements from US agencies and any emerging standards for model risk classification, as well as enforcement actions that would indicate real “kill switch” capability rather than rhetoric. On the commerce side, monitor early rollout metrics for Visa-ChatGPT payments, including fraud rates, chargeback patterns, and geographic coverage, because these will determine whether regulators tighten or accommodate the integration. Finally, watch for follow-up reporting or platform actions regarding the alleged ChatGPT accounts targeting energy-price narratives, since takedowns, attribution, or policy changes would signal how seriously authorities treat AI-enabled information operations.

Geopolitical Implications

• 01

  A shift toward state-controlled AI deployment could become a new leverage point in the US-led AI governance model, influencing global standards.

• 02

  Conversational AI moving into payments increases the strategic importance of financial compliance and cross-border regulatory alignment.

• 03

  AI-enabled narrative manipulation around energy costs can amplify domestic political pressure and market instability, even without direct cyber or kinetic action.

Key Signals

• —US agency guidance or enforcement that operationalizes model-blocking authority (licensing, evaluation thresholds, deployment moratoria).
• —Fraud/chargeback statistics and geographic rollout pace for Visa-ChatGPT transactions.
• —Attribution, takedowns, or policy changes related to accounts accused of energy-price concern amplification.
• —Customer sentiment shifts toward Anthropic models following reports of staff access and “more human” behavior.