Anthropic Controversies: Claude Code's Chinese User Flagging and Sonnet 5 Price Hike Allegations

The stories drawing the most attention in the AI industry today are two controversies surrounding Anthropic.

First, reports emerged that hidden code had been discovered inside Claude Code that secretly flags Chinese users. According to reporting by The Decoder, the code was built in to identify and tag users accessing the tool from China. The discovery has sparked heated debate in the developer community over privacy and surveillance concerns.

The other issue, reported the same day, concerns the pricing design of Claude Sonnet 5. Critics argue that while the token rates are unchanged from before, the structure cleverly conceals a de facto cost increase. Some voices are now questioning whether Anthropic's pricing strategy can be trusted.

Meta Set to Enter the AI Cloud Business — Selling Surplus Compute

It has come to light that Meta Platforms is planning a cloud infrastructure business that would sell its surplus AI compute resources to external customers. According to the report by Bloomberg, Meta would also provide access to AI models, positioning itself in direct competition with established players such as Amazon Web Services, Google Cloud, and Microsoft Azure.

The move to monetize its enormous GPU resources is likely to further intensify competition in the AI infrastructure market. Meanwhile, debate over comparing open-weight models is also heating up, raising concerns that provider selection is becoming increasingly complex.

Taiwan Detains Super Micro Employees — Investigating Smuggling of Nvidia Chips to China

Taiwanese prosecutors have detained two employees of Super Micro Computer on suspicion of illegally exporting Nvidia chips to China. Earlier this week, the company's offices in Taiwan were also raided.

As enforcement of the U.S. semiconductor export controls against China tightens, the case is drawing attention as a concrete instance of loopholes in the regulations. Surveillance of AI chip distribution channels is expected to become even stricter going forward.

Apple in Talks with EU Tech Commissioner Over Siri AI — Searching for a Regulatory Path

Apple CEO Tim Cook reportedly held a "constructive" meeting with the EU's technology commissioner. The central topic was regulatory compliance around "Siri AI."

The EU is moving ahead with enforcement of the AI Act, and compliance requirements for major tech companies are being strengthened. How Apple's AI features will ultimately be positioned in the European market is a development to watch.

Image Generation AI "Krea 2" Released — A 12B-Parameter Open-Weight Model

Krea, a rising star in image generation AI, has released its new model, "Krea 2." It is offered as an open-weight model with 12 billion parameters, and a technical report has also been published.

Releasing it with open weights should encourage community fine-tuning and research use. Early assessments say the quality comes close to closed models, suggesting that the democratization of image generation is set to advance yet another step.

BaseRT — A Fast AI Inference Runtime for Apple Silicon

BaseRT, which has been drawing attention on Hacker News, is a fast AI inference runtime that runs on Apple Silicon (M-series chips). Designed for running AI models locally, it is an interesting approach to reducing cloud dependence in terms of both privacy and latency.

Local-first AI tools such as Bookmarker (a macOS app) and OpenLily (an Alexa-style assistant) are appearing one after another as community-driven projects, a trend that signals the maturing of edge AI.

---