2026-09-20
Bessent Opens New York Talks With China's He Lifeng as US Names Six Chinese AI Firms Over Model Theft, Rare Earth Flows Still 'Not Up to Par'
In this article
What Happened
On Sunday, September 20, Treasury Secretary Scott Bessent and US Trade Representative Jamieson Greer sat down with Chinese Vice Premier He Lifeng at JPMorgan Chase's headquarters in Manhattan. The talks began around 10:30 a.m. Eastern and were expected to run all day, laying the groundwork for Thursday's (September 24) White House summit between President Trump and President Xi Jinping. Xi's trip marks his first state-level visit to Washington in more than a decade, effectively returning the favor for Trump's own state visit to Beijing back in May.
The agenda splits into three tracks. First is AI governance - specifically, allegations that Chinese AI companies have been systematically stealing capabilities from American models. Second is the flow of rare-earth magnets and critical minerals that are essential to advanced semiconductor manufacturing. Third is the status of the US-China tariff truce struck in Busan last October, which is set to expire November 10 - meaning this weekend's talks and Thursday's summit are effectively the last real coordination window before that deadline.
Of the three, the AI theft issue has drawn the sharpest market attention. Two days before the talks, on September 18-19, the NSA, FBI, and CISA issued a joint advisory formally naming six Chinese AI companies - DeepSeek, Moonshot AI (maker of Kimi), Alibaba, MiniMax, StepFun, and Z.AI. The advisory accuses them of running industrial-scale "knowledge distillation" campaigns that systematically extracted billions of tokens from top US models - Claude, GPT, Gemini, and Grok - since late 2024. US officials say the technique effectively lets Chinese firms sidestep American chip export controls by acquiring frontier AI capability without needing the restricted hardware to develop it from scratch.
Just before the talks, on Friday, September 18, a senior US official told reporters that China's progress on restoring rare-earth flows "has not been up to par" - a reference to Beijing's commitment, made as part of last October's Busan truce, to normalize rare-earth exports that it had previously curtailed. That shortfall is expected to be a central sticking point in Sunday's session.
Why This Talks Session Matters Directly to US Markets
The AI theft dispute didn't emerge out of nowhere. On June 24, Anthropic sent a formal letter to the White House and to US senators alleging that Alibaba's Qwen AI lab had run roughly 28.8 million exchanges with Claude models using about 25,000 fraudulent accounts between April 22 and June 5 - what Anthropic called the largest distillation attack it had ever documented. When the news broke on June 25, Alibaba's US-listed ADR (ticker BABA) fell nearly 5% in a single session and lost more than 9% cumulatively over the following five trading days, hitting a 16-month low. The story resurfaced on September 10, when Anthropic disclosed a broader analysis finding that of roughly 200 million suspected distillation exchanges, 151 million traced back to Alibaba's Qwen team - triggering three more consecutive down sessions for the ADR (down 3.9%, then 2.7%, then 4.7%).
Bessent has framed the issue in blunt terms, saying "open source is not open season on American IP," and has warned that sanctions and Entity List designations are "on the table" if the distillation attacks are found to cross into outright IP theft. That framing explains why AI policy is being negotiated at the same table as tariffs and trade this weekend rather than treated as a separate technical matter: from Washington's perspective, chip export controls and AI IP protection are two sides of the same containment strategy.
The stakes extend into export-control policy itself. The rationale behind restricting Nvidia and AMD's advanced AI chip sales to China has always rested on the idea that limiting raw compute would slow China's ability to catch up on frontier AI capability. If the distillation allegations hold up, they suggest Chinese firms have found a way to indirectly replicate the reasoning ability of top US models without needing to acquire large volumes of the restricted chips in the first place - a finding that calls the entire premise of export controls into question. How aggressively the US responds in these talks could therefore move the policy-risk premium priced into China-exposed semiconductor names like Nvidia and AMD, not just AI-adjacent Chinese ADRs. The concern isn't new: when Chinese startup DeepSeek claimed in early 2025 that it had matched top US model performance using a fraction of the compute, some in the industry immediately floated distillation as a possible explanation. This week's joint federal advisory effectively elevates that earlier suspicion into an official, government-documented finding.
Rare-earth stocks have shown just how headline-sensitive this negotiation has become. Throughout September, US-listed rare-earth names have swung sharply on every scrap of news: optimism about a trade thaw tends to sell them off, on the logic that warmer relations would erode the scarcity premium behind the trade, while signs that China is dragging its feet on supply commitments tend to send them sharply higher. That dynamic played out in real time on Friday, September 18, when Bessent's "not up to par" comment hit the wires and MP Materials and USA Rare Earth both jumped roughly 8-9% in a single session - a reminder that these stocks are currently trading more on negotiation headlines than on quarterly fundamentals.
Taken together, Sunday's session sits at the intersection of two opposing forces for US-listed assets. The harder Washington pushes on AI model theft, the more risk premium gets priced into Chinese AI-linked ADRs like Alibaba and, by extension, into the broader basket of US-listed Chinese tech names. Conversely, any sign that Beijing is moving to placate US concerns on rare earths would tend to work against MP Materials and its peers, since it would undercut the scarcity story that has been driving their recent rallies. Two different agenda items, negotiated in the same room on the same day, are capable of pulling different corners of the market in opposite directions.
What to Take Away From This
- A single negotiation can produce winners and losers depending on which agenda item gets emphasized. Escalation on AI theft tends to weigh on China-linked tech ADRs, while any softening on rare-earth supply tends to weigh on US rare-earth miners - so read headlines for which specific track they're addressing, not just the summit as a whole.
- When a government advisory and a private company's own investigation land at the same time, the combined weight tends to escalate an issue faster than either alone. The CISA/NSA/FBI advisory landing alongside Anthropic's own findings turned what had been a company-specific dispute into a bilateral negotiating point within days.
- Stocks caught up in active policy disputes can move more on headlines than on fundamentals in the short run. MP Materials swinging roughly 8% in a single session on a single official's comment is a case study in how policy-sensitive names can decouple from earnings-driven valuation over short windows.
- Negotiations with a hard deadline tend to get more volatile, not less, as the deadline approaches. With the tariff truce expiring November 10 and this week compressing both preparatory talks and the leaders' summit into a few days, both a breakdown and a dramatic breakthrough remain live possibilities that markets will need to price quickly either way.
FAQ
What exactly is AI "distillation," and why is it controversial?
Distillation is a legitimate machine-learning technique where a smaller, cheaper model is trained using the outputs of a larger, more capable one. The controversy arises when it's used adversarially - repeatedly querying a rival's model at scale to harvest its reasoning patterns and effectively clone its capabilities without the underlying research investment. That's the specific misuse US officials and Anthropic allege against the six named Chinese firms.
When will the outcome of Sunday's New York talks show up in stock prices?
The talks themselves are private, and any formal announcement is likelier to come around Thursday's Trump-Xi summit than immediately after this session. That said, markets have shown a pattern this month of pricing in each new headline or official comment in real time - moving Treasury yields, the dollar, China-linked ADRs, semiconductor names, and rare-earth stocks well before any formal deal is announced.
What happens if the tariff truce isn't extended past November 10?
If the two sides fail to extend the truce struck in Busan last October, tariffs and export-control measures that have been paused could snap back into effect. That would directly affect the cost structure and supply chains of US-listed companies with significant China exposure in semiconductors, rare earths, and cross-border trade - which is why markets are treating this week's talks and summit as the clearest early signal of whether that deadline gets pushed back.
For related coverage, see: Week Ahead: Trump-Xi White House Summit Headlines a Week That Also Carries the AI-Slowdown Fight, 'Golden Cross' Trap: Bitcoin Touches $81,700 - Bear-Market Bottom or Short-Squeeze Trap?
Sources
This article is an original synthesis and analysis based on the reporting below, not a reproduction of the original articles. Please check the source articles directly for the most current figures and details.
- US Treasury's Bessent, China's He to launch talks on AI, trade, critical minerals - Reuters (via CNBC)
- CISA, NSA and FBI Warn of China-Based AI Companies Targeting US AI Models with Industrial-Scale Knowledge Distillation Campaigns - CISA
- Anthropic Accuses Alibaba of Massive AI Data Distillation Attack. How to Play BABA Stock Here - Barchart
- US Accuses Six Chinese AI Firms of Copying Models, Bypassing Chip Controls - Tech Times
⚠️ This article is for informational purposes only and is not investment advice. Market conditions change constantly - always verify the latest data before making any investment decisions.