New Delhi: Artificial intelligence (AI) could transform scientific research by accelerating drug discovery and allowing smaller laboratories to tackle complex biological problems. However, the same technology could also make it easier for malicious actors to develop biological weapons, Tencent’s chief AI scientist has warned.
Vinces Yao Shunyu, Tencent’s chief AI scientist and a former OpenAI researcher, said increasingly powerful AI models were approaching a stage where their capabilities could create serious risks in biology and cybersecurity.
Speaking at a science symposium in Hong Kong on Saturday, Yao cautioned that AI systems were moving closer to capabilities that could be misused to cause real-world harm.
“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,” he said.
His warning highlights a growing challenge for the technology industry: how to make advanced AI tools accessible to researchers while preventing their misuse for dangerous activities.
AI offers new possibilities for drug discovery
AI is increasingly being used in scientific research, particularly in areas involving proteins, biological processes and the development of medicines.
Traditional drug discovery can require years of research, significant funding and specialised scientific expertise. Researchers must identify promising biological targets, study molecular interactions and evaluate potential compounds before a medicine can progress through development.
AI models can assist scientists by analysing large volumes of biological data, predicting protein structures and identifying patterns that may be difficult to detect through conventional methods.
These capabilities could help researchers explore new approaches to disease treatment and reduce the time required for certain stages of drug discovery.
The technology could also benefit smaller laboratories and independent scientists that do not have access to the extensive resources available to large pharmaceutical companies and research institutions.
By making advanced computational capabilities more widely available, AI could expand participation in scientific research and allow smaller teams to investigate complex problems.
However, Yao warned that the benefits of wider access must be considered alongside the potential risks.
Why open-weight AI models raise safety concerns
The debate is particularly relevant to open-weight AI models.
Unlike conventional closed AI systems, where the underlying model weights remain under the control of their developers, open-weight models can be downloaded and modified by users with sufficient computing resources.
This flexibility allows researchers and developers to customise models for specialised applications. Independent laboratories can adapt them for scientific investigations, while cybersecurity professionals can use them to identify vulnerabilities and strengthen digital defences.
However, the same accessibility can create opportunities for misuse.
People with malicious intentions could attempt to exploit advanced biological capabilities to design dangerous pathogens or develop biological weapons. As AI systems become more capable, the possibility of using them to assist harmful biological research has become an important concern for scientists and policymakers.
Yao said biological AI capabilities could be more difficult to regulate than cybersecurity applications because defensive and offensive uses are not always easy to distinguish.
A model that helps scientists understand biological processes could potentially provide information that someone else might attempt to misuse.
He stressed that companies developing open AI models must take their security responsibilities seriously.
“All the Chinese or any open model company needs to be very responsible and very serious about those security and safety concerns,” he said.
Tencent explores autonomous AI research laboratories
Tencent is also exploring how AI could become more directly involved in scientific research.
The Chinese technology company is working towards the development of autonomous science laboratories, where AI systems could carry out complex research processes and experiments with limited human intervention.
The proposed approach includes scientific tasks such as protein folding and other complicated experiments that traditionally require specialised researchers and laboratory infrastructure.
The objective is to expand the role of AI beyond answering questions, generating content and assisting with routine tasks. Instead, the technology could potentially help plan and execute scientific investigations from beginning to end.
Such developments could change how research teams operate, allowing scientists to use AI systems for complex analytical and experimental work.
However, the move towards greater autonomy also makes safety evaluations important. Systems that can independently perform scientific tasks need safeguards to prevent their capabilities from being used in harmful ways.
Tencent’s Hy4 model adopts a safety-focused approach
Tencent’s latest flagship AI model, Hy4, reflects the company’s attempt to balance model accessibility with security restrictions.
According to the company, it removed the model’s hacking capabilities while retaining functions that support cybersecurity defence.
Its biological capabilities, meanwhile, remain below the company’s high-risk threshold.
Tencent plans to release the model weights alongside a system card explaining the model’s cybersecurity and biological safety evaluations.
A system card generally provides information about a model’s capabilities, limitations, testing and safety measures. Such documentation can help researchers and users understand the potential risks associated with deploying an AI system.
The planned release of Hy4’s weights will also be closely watched in the wider debate over open AI development, particularly as technology companies face growing pressure to demonstrate how they manage potentially dangerous capabilities.
Open-source or closed AI: Which approach is safer?
Yao does not believe that either completely open or entirely closed AI models necessarily provide the ideal solution.
Open-weight models can encourage innovation by allowing developers and researchers to examine, adapt and deploy AI systems for different purposes. They can also reduce dependence on a small number of technology companies.
Closed models, on the other hand, give developers greater control over access and deployment. This can make it easier to restrict certain capabilities, although it may also limit legitimate research and defensive applications.
Yao pointed to an incident involving Hugging Face, which reportedly used a Chinese open model to defend against a cyberattack after closed US models declined to provide assistance.
The example illustrated how restricting access to powerful AI systems could also create challenges for legitimate cybersecurity work.
“Maybe the best way forward is still a mix of closed and open-source because you still want to empower everyone to leverage the technology,” he said.
His position suggests that the future of AI development may involve different access models depending on the capabilities and risks associated with individual systems.
Balancing scientific progress with biological security
The warning from Tencent’s AI chief comes as the technology industry explores increasingly advanced applications of artificial intelligence.
AI-assisted drug discovery and biological research could create significant opportunities for healthcare, scientific innovation and disease treatment. At the same time, the possibility of misuse has made biological safety an important issue for AI developers.
The challenge is to ensure that researchers can benefit from advanced tools without providing unrestricted access to capabilities that could facilitate dangerous activities.
This will require continued investment in safety testing, risk assessment, access controls and responsible model development. Cooperation between technology companies, scientific institutions and policymakers will also be important as AI capabilities evolve.
Yao’s remarks underline that the debate over open and closed AI models is no longer limited to innovation, competition or commercial interests. It increasingly involves questions of public safety and the responsible use of powerful scientific technology.
As AI becomes more capable of contributing to real-world research, the industry will need to find ways to preserve its scientific benefits while reducing the risks of biological and cybersecurity misuse.