It was a day when the debate over slowing AI development pulled in both diplomacy and the financial markets. In response to U.S. AI executives' calls for a slowdown, China's Foreign Ministry fired back, calling them "fearmongering." In the U.S., AI-related stocks fell, with the slide spreading to Japan's SoftBank, Kioxia, and others. Meanwhile, code suggesting that Apple's Siri can be swapped out for an external LLM has been drawing attention — a reminder that the news isn't all about the slowdown debate. Here's a roundup of today's stories.
China Hits Back at AI Slowdown Warnings as "Fearmongering" — The Standoff Comes Into Focus
After U.S. AI industry leaders, including Anthropic's Dario Amodei, took turns warning of AI risks and urging a slower development pace, China has now pushed back explicitly. According to [The Decoder](https://the-decoder.com/china-fires-back-at-u-s-ai-safety-warnings-calling-them-fearmongering-to-lock-in-american-advantage/), China's Foreign Ministry dismissed the U.S. AI safety warnings as "fearmongering." China's security minister has also come out in favor of accelerating the buildout of AI infrastructure rather than slowing development, and the report notes the view that behind the warnings lies an attempt by the U.S. to lock in its own advantage.
CNBC and Forbes also covered China's "fear mongering" criticism, while the BBC reported that China rejected the accusation that it is engaged in "malicious competition" as well. The slowdown debate is no longer confined to U.S. domestic politics — it is increasingly being read through the lens of U.S.–China relations.
Moves on the U.S. side continue apace. According to [WIRED](https://www.wired.com/story/ai-leaders-are-calling-for-a-slowdown-trumps-team-says-its-on-them/), after Amodei pressed Washington for a development slowdown, Sam Altman and Elon Musk also backed calls for restraint over the weekend. The Trump administration, meanwhile, has signaled that it will leave responsibility for slowing down to the companies themselves, with Trump reportedly prioritizing the maintenance of America's edge over China. Alongside previously reported moves toward concrete measures — such as pre-training safety checks and voluntary coordination among competitors — the tug-of-war over "who sets the pace" looks set to continue.
AI Stocks Plunge in the Wake of the Slowdown Debate
The impact has spread to the markets. The BBC reported that U.S. tech stocks fell as slowdown rhetoric mounts. In Japan, [ITmedia AI+](https://www.itmedia.co.jp/aiplus/article/2609/14/2000001462/) reported that AI-related stocks such as SoftBank, Kioxia, and Tokyo Electron plunged after U.S. AI developers warned that "the pace of development must be slowed to prevent threats to humanity." Safety warnings have now directly shaped investors' risk perception, and volatility is expected to continue as the market watches for regulatory developments and how companies respond.
Apple's Siri Can Be Swapped for Claude or ChatGPT — What the Code Reveals
A [MacRumors report](https://www.macrumors.com/2026/09/14/siri-can-be-swapped-out-for-chatgpt-claude/) is drawing attention on Hacker News. An analysis of Apple's code reportedly shows that Siri's AI features are designed so they can be swapped out for Claude or ChatGPT. How much choice users will actually get remains unclear at this point, but the report stands out as a sign that the industry is moving toward designs that separate the on-device assistant from the backend LLM, making it possible to swap in multiple models.
Over 100 European Politicians Targeted by Deepfake Sites
According to a [WIRED investigation](https://www.wired.com/story/sexually-explicit-deepfake-sites-target-100-plus-politicians-in-europe/), an analysis of 160 websites hosting sexual deepfakes found that more than 100 politicians from 22 countries appear on them — and nearly all are women. The data shows that generative AI abuse is already widely used as a weapon against individual politicians, and it is likely to influence the debate over regulation.
Local LLMs: K2 Horizon 7B Generates Buzz, and a Maple 20B-A1B PR Lands in llama.cpp
In the local LLM scene, the K2 Horizon series (particularly the 7B) is the talk of Reddit's r/LocalLLaMA. [User reports](https://www.reddit.com/r/LocalLLaMA/comments/1wg0vqz/the_new_k2_horizon_models_seem_like_an_absolute/) claim the 7B model significantly outperforms other models at a far smaller size, with some arguing that if it is "genuinely strong rather than benchmark-optimized," it could mark a new milestone. The fact that the entire training process has been made open is also earning praise.
Meanwhile, a pull request has been submitted to [llama.cpp](https://github.com/ggml-org/llama.cpp/pull/27000) to add Maple 20B-A1B, a ternary MoE architecture. With 20B total parameters and 1B active, it is expected to run on CPUs — likely to draw interest from users who want to run large MoE models in VRAM-constrained environments.
Wrapping Up
The slowdown debate has entered a new phase: warning → rebuttal → market reaction. With safety claims now being translated into the language of geopolitics and finance, technical progress — swappable LLMs in Siri, ever-lighter local LLMs — continues in parallel. The next focal points will be how U.S. regulation takes concrete shape, and how China responds in substance.