Oct 25, 2026

US and China AI Blocs Squeeze Allies as Chatbots Pick Sides

Countries are being pushed to choose between US- and China-led AI blocs, while a new study finds Grok and GPT-6.1 lean in opposite directions on US midterm races.

Listen to this update

The global race for artificial intelligence is turning into a contest for allies. A new Associated Press report describes countries being pressed to choose between a US-led AI alliance and a rival group inspired by China, while a study of two leading chatbots found they lean in opposite partisan directions when forced to pick candidates in the US midterms. Sweden, meanwhile, is cutting off popular AI assistants inside its government offices, and China's Moonshot has shipped a model that nearly matches the best US systems on a key coding test.

Two AI blocs, and the countries in between

According to the AP, the US-backed Pax Silica group, which aims to secure AI supply chains, and China's World AI Cooperation Organization (WAICO) are increasingly competing for the same members. WAICO launched with 29 founding countries, among them Pakistan and Russia, and Indonesia has since joined. Singapore says it is weighing membership.

Washington has made clear it sees the two as incompatible. The State Department told the AP that countries pursuing initiatives that promote a competing vision for superintelligence cannot be trusted partners in Pax Silica. China's embassy countered that AI should not belong to major powers alone.

Kazakhstan shows how awkward the squeeze has become. Its AI minister traveled to Washington in June to join Pax Silica, then became a founding WAICO member at a ceremony in Shanghai roughly three weeks later. The country has signed a deal worth up to $10 billion with Nvidia for a planned "data center valley," while Chinese direct investment there reached a record $2.8 billion in 2025. "Everyone is hedging," a former Kazakh diplomat told the AP.

Other governments are trying to avoid the choice altogether. Indonesia's Mari Pangestu said the country wants to belong to both blocs, and Japan, though closely tied to the US on chips and security, is not ruling out cheaper Chinese models for lower-risk uses. Research group AidData has counted Chinese grants and loans for AI-related projects in at least 77 low- and middle-income countries, a sign of how actively Beijing is courting partners.

Chatbots pick opposite sides in the midterms

A study by Convovo, a research studio led by Cornell professor Tapan Parikh, asked two chatbots to name a single candidate in nearly 500 US midterm races, TechCrunch reports. Queried through their developer APIs with a prompt that ruled out hedging, xAI's Grok chose the Republican in 92% of answers, while OpenAI's GPT-6.1 Sol chose the Democrat in 97%. The two models agreed less than 10% of the time. The consumer apps declined to answer the same question.

The models also gathered information differently. Grok often cited X and conservative voter guides alongside nonpartisan sites such as Ballotpedia, while GPT leaned on Wikipedia, the AP, Axios and Reddit. The researchers also found roughly 1,240 extra input tokens attached to every Grok API request that they had not written and could not see.

When TechCrunch asked for a rerun with web search turned off, the picture shifted. GPT still picked the Democrat in 24 of 30 races, but Grok declined to choose in 27 of them. Parikh concluded that GPT's lean is "in the weights," while Grok's depends on what it searches for. OpenAI said the test used a developer API and a forced-choice prompt that does not reflect typical use. Pew data cited in the report shows only 8% of Americans use chatbots for election information.

Sweden blocks ChatGPT and Claude in government offices

Starting October 26, staff at Sweden's Government Offices will no longer be able to use OpenAI's ChatGPT or Anthropic's Claude on work computers, according to Crypto Briefing, citing Sveriges Radio. Microsoft Copilot remains the only AI assistant approved for general use, and it cleared an internal security review only after its settings were adjusted. A small group of staff will keep access to both blocked models through an in-house platform.

The reason given is information security, particularly the use of AI services that run without a formal contract. "We have seen several examples of problematic use of open AI services," said digitalisation director Mikael Wettercrantz. The Swedish Tax Agency and the National Board of Institutional Care have adopted similar limits.

Moonshot's Kimi K3.1 Pro closes in on the leaders

China's Moonshot AI released Kimi K3.1 Pro on October 23. The company reports a score of 65.7% on Terminal-Bench 4.0, a coding-agent benchmark, just behind Claude Opus 5.5 at 66.4%. Independent evaluator Artificial Analysis ranks it fourth on its Intelligence Index with a score of 55, an 11-point jump from Kimi K3, and says its API prices are a third lower than its predecessor's. Moonshot plans to publish the full model weights by November 4, at which point Artificial Analysis expects it to become the highest-scoring open-weights model.

Why it matters

Taken together, these stories show AI becoming a question of trust at every level. Governments are deciding which bloc to rely on, agencies are deciding which tools staff may use, and voters may soon need to ask which way the assistant in their pocket leans.