Image from Chosun titled “Anthropic logo /Reuters-Yonhap”
In June, Anthropic, the AI lab behind the Claude chatbot and various agentic coding tools, told the US Senate that AI labs affiliated with Qwen and China were doing distillation attacks on Anthropic’s models. Then in July, it got caught doing something worse: quietly spying on the users it claimed to protect.
On June 30, 2026, a Reddit user reverse-engineered the Claude Code software and found something strange. Since version 2.1.91 came out on April 2, 2026, Claude Code had been checking in the background a user’s timezone and internet connection, and if it looked like it was coming from China, the tool never said a word. Instead, it changed its output, tweaking date formatting and punctuation to send a hidden message to Anthropic’s servers. This was just a silent report without a notice, warning, consent, or anything like that.
After this report was made on Reddit, Thariq Shihipar, an engineer on the Anthropic Claude Code team, confirmed the discovery was real. He framed it in terms of user security, saying that the code was “meant to prevent account abuse from unauthorized resellers and protect against distillation.” Distillation is simply the practice of training a cheaper AI off Claude’s answers without permission. He said Anthropic planned to remove it.
But that didn’t erase the irony of this whole situation. Three weeks earlier, Anthropic told the US Senate how operators linked to Alibaba’s Qwen lab, which is based in China, ran the largest distillation attack against Claude to date. Anthropic said operators used about 25,000 fraudulent accounts and exchanged about 28 million messages across six weeks. Anthropic built the secret code to catch a behavior it was accusing China of.
Alibaba then banned the usage of Claude Code across its entire workforce, requiring all of its employees to use AI tools built by the company instead. The defense Anthropic built to protect itself from Chinese competitors ended up handing away one of the biggest tech companies on Earth a reason to walk away from them.
And the direct trouble with China isn’t the only problem Anthropic is facing. The indirect consequences of this can be seen with the recent Fable 5 launch. Fable 5 is Anthropic’s flagship model, which has undergone significant scrutiny from the US government. Fable 5 has spent the last month going on and off. It launched on June 9, 2026, got pulled away 3 days later under US export controls, came back July 1, 2026, and today, July 12, 2026, is the deadline before it stops being included in Pro and Max subscriptions. However, it’s been extended to July 19, 2026. Meanwhile, China might lock down its own models using the same national security logic Washington used against Anthropic.
Chinese models now make up about 30 to 46 percent of enterprise AI models used by companies in the USA, up an average of 11 percent since last year, and none of this volatility is slowing down this shift to Chinese AI tools. These enterprise companies aren’t switching to Chinese AI products because they’re better, but because they’re cheaper and compare directly with contemporary, US flagship models.
Every move meant to protect the company’s turf- tracking code, export bans, Senate letters- is pushing the other side farther away and growing discontent among the public, for whom AI is being improved. Nobody is racing towards a common goal finish line anymore. Both sides are just building higher walls.
Be First to Comment