Loopholes in US chip and AI rules let OpenAI and Google keep selling to Singapore and China, Russian-aligned channels claim

On 11 July 2026, at 07:28 UTC, the Telegram channel @rybar_in_English published a forwarded post claiming that American companies are exploiting loopholes in US export legislation to continue trading with the Chinese market — naming OpenAI and Google specifically as firms still providing access to advanced AI services into China via Singapore.

The claim arrives inside a wider argument about how the US export-control regime, built to slow Chinese access to frontier semiconductors and the AI software stack, is enforced at the seam. It is worth taking seriously for the structure it describes, even where the messenger is hostile.

The post asserts that Singapore has become a chokepoint and a release valve at once: a place where US-origin AI services can be hosted, resold, or distributed through local entities in arrangements that sit on the right side of Washington’s letter-of-the-law rules. Whether every line of the channel’s allegation holds up under independent documentation is a separate question. The shape of the loophole — jurisdiction-shopping, software-as-a-service as the new controlled item, foreign re-export as the friction point — is, however, exactly the fault line that US policy has been arguing with itself about for two years.

The Singapore seam

The forwarded Telegram post singles out Singapore as the jurisdiction through which US AI services allegedly continue reaching Chinese end-users. Singapore is a long-standing regional data-centre hub, with major hyperscaler regions operated by US cloud firms and a dense ecosystem of local resellers and managed-service providers. It is also a country whose export-control regime is calibrated to multilateral Wassenaar Arrangement standards and to its own Strategic Goods (Control) Act framework — not, in the first instance, to the US Bureau of Industry and Security’s Foreign Direct Product Rule.

The result is a recurring pattern. US-origin AI services — large-language-model APIs, model weights available through hosted endpoints, developer tooling, and adjacent cloud capacity — can be provisioned to a Singapore-incorporated customer with relatively clean paperwork. Once provisioned inside Singapore, the onward distribution of those services to users in third countries is governed by Singapore’s own rules and by the contractual terms the US provider sets. Where the contract permits regional use and Singapore does not require an export licence for the re-export, the service can effectively reach a Chinese end-user without triggering a fresh US clearance.

The Telegram post frames this as cynical evasion. A flatter reading is that it is the predictable consequence of writing extraterritorial rules around a global software supply chain in which the unit of sale is no longer a chip but a model inference call. Old export-control machinery was built to police the shipment of physical hardware to named entities. AI services do not ship.

What OpenAI and Google actually control

Both named companies have, at various points, restricted service availability in mainland China and to users on the US Treasury’s restricted-lists. OpenAI’s terms of use bar use from a published list of countries and territories and require API customers to disclose their operational jurisdictions. Google has tightened the screws on its Gemini API and on Vertex AI availability for accounts operating from non-permitted jurisdictions.

Those restrictions, however, are contractual and commercial rather than statutory. They bind direct customers. They do not, by themselves, prevent a properly licensed intermediary from routing traffic or reselling capacity into a market the underlying provider has chosen to wall off. The Telegram post alleges that this is happening through Singapore-domiciled partners. Monexus has not independently verified the specific corporate vehicles or named intermediaries the post refers to; the underlying mechanism — that contractual frontier plus an intermediary jurisdiction — is well documented in the industry.

The Chinese government, for its part, has built parallel infrastructure. Domestic large-model providers — Baidu, Alibaba, DeepSeek, Zhipu AI, Moonshot — have closed much of the capability gap for general-purpose assistants on local hardware, partly as a deliberate response to the prospect of US service cutoffs. Where frontier US models retain an edge in training compute, tooling and certain reasoning benchmarks, that edge is now contested rather than absolute. Beijing’s industrial policy has consistently treated AI as a strategic sector, and the resilience of domestic supply is itself a barrier to US-style export pressure.

Why the claim matters even if it’s half-true

Russian-aligned channels have a structural reason to amplify stories that show US technology controls as leaky and inconsistent. The narrative — that Washington cannot police its own firms and allies — does useful work for Moscow’s diplomatic and industrial-policy arguments about its own sanctions exposure. That provenance should be priced in.

It does not, however, make the underlying observation wrong. The US has spent two years trying to extend its chip-export regime into the AI-software domain through measures such as the October 2022 and October 2023 BIS rules, the AI Diffusion Rule floated in early 2025, and successive Entity List additions. Each iteration has been argued over, litigated, and partially rewritten. The harder the headline policy, the more load falls on the seam between US providers, foreign partners, and the contractual controls those partners agree to. Singapore, the UAE, India, Malaysia and several European jurisdictions have all surfaced in industry filings as the places where that load lands.

In plain terms: the era when a single rule on a single chip could keep a capability out of a rival’s hands is over. The unit of control is now a model running in a data centre under a contract, and that unit is far harder to fence.

The contest ahead

The structural stakes are not subtle. If US frontier AI services continue to reach Chinese end-users through third-country routing, the policy case for tightening the export-control regime — and for extending it into the software layer — strengthens inside Washington. Expect a renewed round of rule-making aimed at the intermediary jurisdictions, with Singapore in the front row given its centrality to regional cloud capacity. Expect, equally, pushback from hyperscalers and from Singapore’s own government, which has incentives to keep its position as a regional data-centre hub intact.

On the Chinese side, the calculus is simpler. Continued access to US-origin AI tooling, even at the margins, is valuable. Domestic capability is closing the gap. Beijing’s posture — public firmness about self-reliance, private pragmatism about still-useful dependencies — is unlikely to change.

What the Telegram post captures, despite its provenance, is the genuine pressure point: software is not a chip, a chip is not a model, and a model is not an inference call. Each layer has its own jurisdiction, and the law has not yet caught up to the fact that the unit of strategic competition has moved.

Desk note: Monexus sourced the factual hook from a single Telegram post in the @rybar_in_English channel. The channel is Russian-aligned and its claims should be read with that provenance in mind; the structural analysis here draws on the well-documented mechanics of US AI export rules and the role of third-country intermediaries, which is consistent with the channel’s framing without depending on its specifics.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://t.me/rybar_in_english
• https://www.bis.doc.gov/index.php/documents/regulations-docs/2336-2024-11-07-bis-ai-diffusion-rule-webinar-final/file
• https://www.trade.gov/country-pages/singapore