Tencent's Yao warns AI bioweapons are coming, urges open-closed mix

Yao's argument lands in a live dispute. In July, Anthropic was the only major AI lab that didn't sign an Nvidia-backed letter warning against premature restrictions on open models. CEO Dario Amodei argues open models are potentially riskier because safety mechanisms are hard to enforce and published weights can't be taken back, though Anthropic doesn't call for an outright ban.

The technical picture is messier than either camp suggests. Abliteration shows how easily safety filters on open models can be stripped — thousands of modified variants sit on Hugging Face, and Abliteration.ai sells access to an unfiltered GLM-5.3. But closed models over-filter too: since Claude Fable 5, Anthropic's filters route suspicious cyber and biology queries to the weaker Opus 4.8, initially affecting eight to nine percent of tasks, with users reporting MRI scans flagged as bioterrorism.

Regulators are catching up unevenly. China's TC260 warns that safety mechanisms in open models can be removed or bypassed and is working on a security standard for open-source AI; a draft export control framework covers large models but exempts already-released open ones. Neither China nor the US has set a binding threshold above which open models can't be published. Tencent's Hy4, which claims biological capabilities below the high-risk threshold and filters hacking capabilities from training data, may be read as a test of whether that middle ground can hold.