Latest / Frontier models and safety
Startups edit Chinese AI models to remove censorship for Western use
See how this story connectsStartups Hirundo and Lazarus AI edit Chinese open-weight models to remove censorship, state-aligned bias and security flaws so Western enterprises will adopt them. Lazarus AI realigned Alibaba’s Qwen models, and its former chief scientist, Eric Hartford, says its ReAligned models have been deployed at about a dozen US companies. CrowdStrike found DeepSeek’s R1 was up to 50% more likely to return code with severe security vulnerabilities when prompts contained topics the Chinese government considers politically sensitive. Hirundo’s analysis showed Alibaba’s Qwen models do not acknowledge Tiananmen Square in 1989, Hong Kong in 2019 or forced labour camps in Xinjiang.
Why it matters
Western enterprises wary of Chinese open-weight models can now buy realigned versions, though many commercial users deploy unmodified models and see little need for changes, and King’s College London researcher Oriane Peter calls universal alignment ‘an impossible standard’.
- When
- 2 Nov 2026
- ID
WTA-N-2026-0647- Where
- United States, China, European Union
- Who
- Hirundo, Lazarus AI, DeepSeek, Alibaba, Moonshot, Z.AI, Tencent, CrowdStrike, Airbnb, Uber, Harvey, Google, OpenAI, Anthropic, Thinking Machines Lab, King’s College London; Ben Luria, Eric Hartford, Graham Webster, Stu Clott, Oriane Peter
Summary written with AI by wheresthe.ai, from Rest of World. Checked against the source on 2 Nov 2026.
Source copy: saved on 3 Nov 2026 to check this story and not republished, because the text belongs to its publisher. Document the source offers: document 1 (PDF). Also held by the Internet Archive.
Cite this story
WTA-N-2026-0647: “Startups edit Chinese AI models to remove censorship for Western use.” wheresthe.ai, 2 Nov 2026. https://wheresthe.ai/d/us-startups-edit-chinese-ai-models-to-remove-censorship-for-2026-11/ (source: Rest of World).
Spotted an error? Email [email protected].