Tencent's detailed model safety card reveals work with startup Fangcun AI on misalignment evals
terryyuezhuo · x · 2026-11-20
terryyuezhuo adds two points about Tencent's safety card: Tencent has a dedicated team working on safety and security evals, and includes people from a Tsinghua network well connected with Fangcun.
Context: @vincechow1 calls it the most detailed model safety card from a Chinese tech company — not standard HarmBench, but detailed discussion of frontier issues from open-weight safety to bio risk to loss of control. Tencent says it partnered with little-known startup Fangcun AI to test on their internal misalignment benchmarks; founder Luke Li Xiaojian gave an interview profiling the company.
More from Companies & People
- Anthropic at $1.5T Valuation: Why This Company Could Only Emerge in Silicon Valley — FuSheng_0306 · 2026-11-20
- OpenAI co-hosts Hacktivate AI 2.0 policy hackathon in Brussels on Dec 1 — MartinSignoux · 2026-11-20
- Databricks CEO: we hit AGI last year by the old definition, then moved goalposts — VraserX · 2026-11-20
- Indian founder deletes rage-tweet and apologizes, praised for maturity as users hail Chinese AI community — itsOmSarraf_ · 2026-11-20
- Chinese AI labs' compute anxiety: Qwen team says under 20% odds of catching OpenAI — Boring_Aioli7916 · 2026-11-20
- Hard Fork on GTA VI and the AI Era; Anthropic Goes Public at $2 Trillion — Hard Fork (NYT) · 2026-11-20