TechWithAdit (TWA)

The US-China AI war just hit a new level. The White House has accused Moonshot AI of covertly distilling Anthropic's Fable 5 to build Kimi K3, and Treasury Secretary Scott Bessent has threatened sanctions and Entity List blacklisting. If the accusation stands, the Chinese AI open-source model boom becomes a trade war trigger. On the same day, Uber made the employment trade explicit: 10% of customer service staff gone, replaced by AI — the first time the company has directly linked layoffs to its AI strategy.

On capabilities, an AI system just scored a perfect 100 at the world's most prestigious mathematics competition. Google's cloud boom is giving Sundar Pichai ammunition to push back on claims it is losing the AI race. And India announced government backing for 20 domestic sovereign AI models — New Delhi's most ambitious AI sovereignty play yet.

Uber announced on Wednesday that it cut 10% of roles in its community operations team — the division handling customer support — as part of a broader move to "simplify operations, strengthen in-person collaboration, and continue to embrace AI," per an Uber spokesperson. VP of Global Community Operations Megha Yethatika told staff in a memo that the organisation had become "too complex and siloed" — making it harder to fully leverage next-generation AI capabilities. Remote workers in the affected team are being required to relocate to hub offices. This is the first time Uber has explicitly linked a headcount reduction to its AI efficiency strategy, following a June incident in which its CTO revealed the company had exceeded its 2026 AI budget within four months.

The Uber cut is not isolated. More than 116,000 tech employees globally lost jobs in the first five months of 2026, with AI cited in the vast majority of announcements. Customer service — repetitive, text-based, high-volume — is the function most immediately vulnerable to AI substitution, and Uber's move signals that the automation is no longer theoretical at consumer-facing scale.

The insights: Uber just gave every customer service department in every company a new benchmark question: if Uber's AI can handle 10% more volume, what percentage can yours? For operators running support teams, the economic pressure to answer that question is now on the clock.

An AI system has achieved a perfect score of 100 at the International Mathematical Olympiad — the world's most prestigious and difficult mathematics competition, which has historically been the exclusive domain of the most gifted human mathematicians. The IMO tests not just calculation but deep mathematical reasoning, proof construction, and multi-step creative problem-solving across six problems over two days. Previous AI attempts had shown partial success on individual problems. A clean 100 is a different category of result. It represents AI demonstrating the kind of structured, creative, long-horizon reasoning that was considered distinctly human territory until this year.

The insights: The IMO perfect score is the reasoning equivalent of AlphaGo beating Lee Sedol — a single landmark that changes the intuition about what AI can do. For operators thinking about AI in roles that require complex structured thinking, the question is no longer whether AI can do the work. It is how quickly the tools reach the same level on the specific reasoning tasks your business actually needs.

Google reported Q2 2026 results that gave Sundar Pichai a strong rebuttal to months of narrative about the company losing ground in AI. Google Cloud grew 31% year-on-year to $13.2 billion in revenue, contributing to operating profit of over $2.5 billion in the division. Pichai told analysts the numbers justify the company's massive AI infrastructure spend — explicitly pushing back on claims that Google was being outpaced by OpenAI, Anthropic, and Microsoft in enterprise AI. The quarter showed Gemini API calls growing significantly, enterprise Gemini seats expanding, and Google Workspace AI features being adopted at scale.

The context that sharpens the story: Google's AI infrastructure commitments exceed $75 billion in planned capex for 2026. Critics have questioned whether the returns justify that spend. Q2's cloud performance is the first clean quarter of evidence that answers the question affirmatively.

The insights: Google's cloud is the revenue engine that makes its AI bet sustainable in a way that pure-play AI labs cannot replicate. For operators choosing cloud AI vendors, Google just gave its enterprise sales team a powerful new closing argument: our AI spending is backed by a business that is actually growing fast.

The Indian government has announced backing for 20 sovereign AI models to be developed domestically, in what marks New Delhi's most ambitious AI self-sufficiency play to date. The models will span multiple sectors — healthcare, agriculture, legal, financial services, and regional languages — and will be developed with government funding, compute access through IndiaAI infrastructure, and data partnerships with public sector institutions. The move follows India's frustration with dependence on US and Chinese AI models for sensitive government and enterprise use cases, particularly after the Fable 5 export ban episode in June demonstrated how quickly access to foreign AI can be revoked.

The insights: India is betting that 20 sector-specific sovereign models — rather than one general-purpose one — is the right architecture for a country with 22 official languages and wildly diverse use cases. For Indian operators, government-backed models with public data access could significantly lower the cost of building AI into regulated and language-sensitive workflows.

Operator's Spotlight Read

On Wednesday, Michael Kratsios, the White House director of science and technology policy, posted on X: "We have information that Moonshot AI distilled Anthropic's Fable for the development of its K3 model. To do this they developed a sophisticated internal platform to conduct large-scale distillation against US models." Hours later, Treasury Secretary Scott Bessent followed with an explicit threat: "Open source is not open season on American IP. When firms conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table." The White House's India Gazette report added a second accusation: Moonshot also allegedly acquired Nvidia GB300 servers through Thailand to circumvent US export controls. Moonshot has not responded publicly. Beijing called the accusations groundless.

The technical dispute is significant and must be stated clearly. Fable 5 was only publicly available from July 1 — less than three weeks before Kimi K3 launched. AI researchers and independent experts have publicly questioned whether large-scale distillation from a model in that timeframe is technically plausible, or whether Kimi K3's capabilities reflect independent development that simply converged at the same capability level. The US government has not released the underlying evidence. The accusation arrived before the proof.

What is not in dispute is the policy consequence. The Trump administration formalised adversarial distillation as a national security threat in April 2026 via NSTM-4. Bessent's threat activates that framework. If Moonshot is sanctioned or Entity Listed, it would become the first Chinese AI company targeted not for hardware violations but for the content of its AI model — specifically, for building a model that outcompetes US frontier labs at 70% lower cost per token. Kimi K3 at $3 per million input tokens versus Fable's $10 is the underlying economic threat. The distillation accusation is the legal mechanism being deployed to address it.

The insights: Whether or not Moonshot distilled Fable, the US government is now establishing a framework in which any Chinese AI model that performs suspiciously well on US frontier benchmarks can be accused of IP theft and sanctioned accordingly. That is an extraordinary expansion of trade enforcement into AI model capabilities. For operators using Chinese AI tools — or building products on open-weight Chinese models — the legal environment around those tools is materially less stable today than it was 48 hours ago.

Was this email forwarded to you? Don't miss any updates — Subscribe to TechWithAdit for sharp, no-noise tech intelligence. Stay sharp. — Adit