Artificial intelligence models will soon be powerful enough to master biology, presenting benefits and serious risks in equal measure, Tencent's chief AI scientist warned during a rare public appearance in Hong Kong on Saturday, adding his voice to a growing debate about what happens when AI systems fully absorb the life sciences.
"The cyber abilities of AI models — and soon the bio ability of models — will reach a threshold where it's actually possible to do harm," said Yao Shunyu, Tencent Holdings' chief AI scientist and a former OpenAI researcher, at a science symposium co-hosted by the Hong Kong Academy of Sciences and the Future Science Prize, according to the South China Morning Post. For more context on this story, see our ongoing AI news.
An Open-Weight Advocate Calls for Responsibility
Yao's warning is notable less for its content than for its source. He built his reputation at OpenAI before joining Tencent, one of China's largest technology companies, and Tencent has positioned itself among the most aggressive publishers of open-weight AI models — systems whose underlying parameters are released publicly and can be downloaded, modified and run by anyone.
That makes his accompanying message pointed: "All the Chinese or any open model company needs to be very responsible and very serious about those security and safety concerns," Yao said, according to the SCMP.
The assessment comes as frontier AI systems are increasingly applied to medicine and biology, where they could accelerate the discovery of new drugs and other treatments — a shift already reshaping China's pharmaceutical industry, according to the SCMP's reporting. Security experts, however, have long warned that the same capabilities could help malicious actors produce bioweapons or spark the next pandemic — a risk that becomes harder to police when capable models are released as open-weight software that anyone can run without guardrails.
Biology is widely viewed among AI experts as the next major frontier after coding, a domain where rapid AI progress has already upended global cybersecurity. Like cyber capabilities, biological capabilities are dual-use: the same model that designs promising drug candidates could, in principle, assist an attacker. There is genuine debate about which side benefits more.
Tencent's Own Safety Playbook
Yao's remarks carried extra weight because they arrived days after his own company shipped a new flagship model. Tencent released Hy4 on Wednesday, which ranked as the second-best performing model produced in China according to benchmarking firm Artificial Analysis, the SCMP reported.
In its release announcement, Tencent described the safety measures built into Hy4. The company said it removed the model's cyber hacking capabilities while retaining its cyber defence capabilities by filtering out relevant data during the training stage. On the biological side, Tencent said Hy4's bio capabilities remained "well below" its high-risk threshold because the model still could not turn foundational biological knowledge into actionable protocols — the gap, in other words, between textbook understanding and step-by-step operational guidance.
That framing — measuring whether a model can convert general knowledge into usable procedures — has become a common yardstick in biosecurity evaluations of AI systems. Tencent's claim that Hy4 falls short of that line is an assertion by the model's own maker, and external validation of such evaluations remains limited across the industry.
The timing also reflects competitive pressure. Tencent unveiled Hy4 into an increasingly heated open-weight race among China's AI firms, with DeepSeek, Z.ai and Alibaba all shipping new models in recent weeks, according to the SCMP's earlier coverage of the launch. In an ecosystem where publishing weights is a competitive necessity, Yao's call for "very responsible" release practices reads as an attempt to set norms inside the very movement that makes safety reviews hardest to enforce.
From Warnings to Autonomous Labs
What makes Yao's appearance particularly interesting is that Tencent is simultaneously pushing AI deeper into the laboratory. Addressing a room of computer science students, he outlined plans for "autonomous science labs" that would use AI systems to run complex experiments end-to-end, with inverse protein folding cited as an example of the kind of work the company is targeting.
Yao argued that the technical obstacles to safe, capable scientific AI would fall in time. "I think all the technical changes can be solved — where it's just a matter of time," he said, per the SCMP. "I think the bigger challenge is societal and organisational."
The venue underscored the message's intended audience: Yao spoke to a room packed with computer science students — the population most likely to build on these models — during the 2026 Future Science Prize Week, which the SCMP reported drew around 1,000 participants.
That division — solvable engineering problems versus unsolved governance problems — is quietly becoming the standard position of AI executives on both sides of the Pacific. For Tencent, it is also a defensive one: a company betting its next act on autonomous laboratories has every incentive to convince regulators that bio-risk thresholds can be engineered, audited and respected in the open.
For now, the near-term question is whether companies publishing open-weight models will adopt measurable bio-risk thresholds before capabilities cross them — or after. Yao's warning suggests Tencent, at least, believes the clock is running.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
