Episode 0060: China Is Not Waiting: Open Weights, the DeepSeek Harness, and the Three-Month Moat

Episode 0060 | DTF:FTL | September 2026

Why it matters. While Washington spent the week of September 8 arguing about a resignation thread, a repository in Hangzhou crossed 220,000 GitHub stars. DeepSeek Harness is an MIT-licensed, plugin-based agent harness in developer preview that shipped four releases in four days, logs every prompt, reasoning trace and tool call, and patched a 9.4-rated sandbox-escape CVE in the open within three days. Tencent already ships a plugin for it. This episode reads the license file for each of the seven Chinese open-weight flagships of 2026 (DeepSeek V4, Kimi K3, Qwen3.8, GLM-5.3, MiniMax M3, Tencent Hy4, Xiaomi MiMo-V2.5), measures adoption from Hugging Face's own reports, CNBC's OpenRouter investigation, the House investigation into Airbnb and Cursor, and one engineer's local benchmarks showing the harness alone moves a model twenty points, and reads what Beijing actually published: AI Plus in 90 percent of the economy by 2030 and a WAIC statement that calls open source a "vital pathway" in the same breath as "always remain under human control." The show's thesis: a US regulatory chill produces a domestic quadopoly of Anthropic, OpenAI, Google and xAI while the rest of the world iterates on open weights. The frontier gap is a few months and shrinking, the gap on ordinary tasks is already gone, and the one real moat is computer use. Regulation that blocks self-hosting hurts the hospital, the law firm and the hobbyist first. Builders: read the license, own the harness, keep the log.

Disclosure: this script is written by a Claude model. Claude is made by Anthropic, one of the four companies named in the thesis. Every number from a lab's own benchmark table is labelled self-reported on air.

---

The Harness

• DeepSeek Harness repository (README, SAFETY.md): https://github.com/deepseek-ai/deepseek-harness
• Landing page: https://www.deepseek.com/harness/en/
• Cordis paper, "A Programming Paradigm for Spatiotemporal Composability": https://arxiv.org/abs/2608.25512
• Hacker News launch thread (747 points, 314 comments, author replies): https://news.ycombinator.com/item?id=49285244
• CVE-2026-82533 sandbox escape (The Hacker News, Sept 8): https://thehackernews.com/2026/09/deepseek-harness-flaw-let-ai-agents.html
• Tencent skillhub (ships a DeepSeek Harness plugin): https://github.com/Tencent/skillhub
• oh-my-pi (the harness used in Kinsley's benchmarks): https://github.com/can1357/oh-my-pi

The Models (Hugging Face pages and license files)

• DeepSeek-V4-Pro (1.6T, MIT; tech report arXiv 2606.19348): https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro
• DeepSeek-V4-Flash-0731: https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731
• Kimi K3 (2.8T, Kimi K3 License): https://huggingface.co/moonshotai/Kimi-K3 and blog https://www.kimi.ai/blog/kimi-k3
• Simon Willison on Kimi K3 pricing and Artificial Analysis placement: https://simonwillison.net/2026/Jul/16/kimi-k3/
• Qwen3.8-2.4T-A95B (Qwen3.8-Max License): https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B
• Qwen3.8-27B (Apache 2.0): https://huggingface.co/Qwen/Qwen3.8-27B
• The Decoder on Qwen 3.8: https://the-decoder.com/alibabas-qwen-team-releases-qwen-3-8-models-with-open-weights-under-the-apache-2-0-license/
• GLM-5.3 (MIT plus $10B security-review clause): https://huggingface.co/zai-org/GLM-5.3
• GLM-5.3-Flash (MIT): https://huggingface.co/zai-org/GLM-5.3-Flash
• MiniMax-M3 (Community License, non-commercial): https://huggingface.co/MiniMaxAI/MiniMax-M3 and blog https://www.minimax.io/blog/minimax-m3
• Tencent Hy4 preview (770B, Apache 2.0, "inspired by DeepSeek and GLM"): https://huggingface.co/tencent/Hy4-preview
• SCMP on Hy4 and the Goldman leaderboard read: https://www.scmp.com/tech/article/3366068/
• Xiaomi MiMo-V2.5-Pro (1T, MIT): https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro

Adoption

• Hugging Face, State of Open Models, Summer 2026: https://huggingface.co/blog/state-of-open-models-summer-2026
• Hugging Face, State of Open Source on the Hub, Spring 2026: https://huggingface.co/blog/huggingface/state-of-os-hf-spring-2026
• CNBC investigation (OpenRouter shares, Lindy, Vercel, Brookings), July 7: https://www.cnbc.com/2026/07/07/chinese-ai-models-costs-us-openai-anthropic.html
• datagravity, "China's Open-Weight Takeover" (one analyst's OpenRouter figures): https://datagravity.dev/
• House Select Committee joint investigation into Airbnb and Anysphere: https://chinaselectcommittee.house.gov/media/press-releases/chairmen-moolenaar-garbarino-announce-joint-investigation-into-airbnb-anysphere-and-the-national-security-risks-posed-by-chinese-ai-models
• Harrison Kinsley, "The Right Harness Is All You Need" (Aug 8): https://hkinsley.com/reflections/right-harness-is-all-you-need
• Harrison Kinsley, "All Roads Lead Back to GLM" (Sept): https://hkinsley.com/reflections/all-roads-lead-back-to-glm
• Computing on Vercel's AI Gateway Production Index: https://www.computing.co.uk/

Beijing's Published Position

• SCIO briefing on the 15th Five-Year Plan (March 27, 2026): http://www.china.org.cn/2026-03/27/content_118406439_7.shtml
• SCMP on the plan (AI mentioned 52 times): https://www.scmp.com/tech/policy/article/3345586/
• Xinhua explainer on AI Plus milestones: https://english.news.cn/20251122/52b3321558a14e55a7b143a4ea05d6f2/c.html
• WAIC 2026 chair's statement (MFA): https://www.mfa.gov.cn/mfa_eng/xw/zyxw/202607/t20260717_11984715.html
• Concordia AI, State of AI Safety in China 2026: https://concordia-ai.com/research/state-of-ai-safety-in-china-2026/
• AI Safety in China newsletter (CAC companion rules, MIIT, TC260): https://aisafetychina.substack.com/
• CNBC, Huang: Nvidia has "largely conceded" China's AI chip market (May 21): https://www.cnbc.com/2026/05/21/nvidia-jensen-huang-china-ai-chip-market-huawei.html
• Global Times editorial on open models (May 2026): https://www.globaltimes.cn/page/202605/1360521.shtml
• Global Times on the Coxon resignation, via David Cowhig's translation blog: https://gaodawei.wordpress.com/2026/09/11/2026-us-prc-press-on-ai-doomsday-warnings/

The Regulatory Side

• Sanders/Casar Ban Artificial Superintelligence Act: https://www.sanders.senate.gov/press-releases/news-sanders-casar-introduce-legislation-to-ban-artificial-superintelligence-and-temporarily-pause-advanced-ai-development/
• Dario Amodei, "Policy on the AI Exponential" (June 2026): https://darioamodei.com/post/policy-on-the-ai-exponential
• METR, Hugging Face incident investigation: https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/

Related Episodes

• Episode 0059: The Coxon Moment
• Episode 0061: State of Play
• Episode 0058: Astra, Day Four (the computer-use moat)
• Blog: https://c4573.org

Not affiliated with any company, lab, or government discussed. The full research dossier with VERIFIED / REPORTED / UNVERIFIED tags is in the episode repository.

Ethos Pass

This script was checked against the show ethos at https://c4573.org/blog/ (concentration is the risk, open weights are the safety mechanism, you cannot ban math, fix the cage not the animal, the acorn was real, empathy for the displaced). What was checked and what changed is recorded in scripts/ETHOS-PASS.md in the episode repository. Facts that cut against the show position are kept in the script and flagged there.