An international group of scientists is calling for stronger responsibility around AI-assisted protein design, warning that powerful life sciences tools could also create biosecurity risks if misused.
The concern is not framed as a rejection of AI. The signatories argue that AI can support valuable work in areas such as infectious disease response and energy production. Their message is that the field needs practical safeguards before dangerous uses become easier to attempt.
What the scientists are warning about
The group released an open letter on the potential of AI in the life sciences. It points to clear benefits, including faster responses to infectious diseases and possible contributions to energy production.
At the same time, the letter warns that AI-assisted protein design could be misused. The risks named in the source include biological toxins and bioweapons.
The current 131 signatories include Nobel laureate Frances Arnold and Turing Prize winner Yann LeCun. They are calling for proactive risk management rather than waiting for misuse to occur.
The core issue is dual use. A technology that helps scientists design useful proteins could also be directed toward harmful biological functions. That does not mean the technology is inherently dangerous in every setting, but it does mean the surrounding research practices matter.
The commitment they want researchers to make
The group pledges to conduct research only for the benefit of society and to avoid dangerous practices. That commitment includes looking beyond the software itself and paying attention to the infrastructure that turns designs into real materials.
One major recommendation is that DNA synthesis services should be procured only from providers that perform standardized biosafety testing. In plain terms, the scientists want researchers to use suppliers that check whether orders raise safety concerns.
The letter also calls for continuous evaluation of AI software and the identification of safety risks. That means risk review should not be treated as a one-time step before launch. As systems change, their safety profile may need to be reviewed again.
The commitments described in the source include:
- Using AI-assisted protein design for the benefit of society.
- Avoiding dangerous research practices.
- Choosing DNA synthesis providers that perform standardized biosafety testing.
- Continuously evaluating AI software for safety risks.
- Reviewing and developing principles over time.
Why DNA production is the key checkpoint
The scientists make an important distinction: they do not describe AI alone as the main point of danger. They argue that real-world harm depends on the physical production of new genetic material.
That is why DNA production facilities become central to the biosecurity discussion. A computational design remains digital unless it is manufactured. The production stage is therefore a practical place to identify and block harmful materials.
"Since no computationally designed protein can cause real-world harm unless it is physically produced, the manufacturing of synthetic DNA presents a key biosecurity checkpoint for the field of computational protein design," the researchers write.
This framing matters because it shifts the safety conversation from general fear of AI to specific controls. Instead of treating every model or every researcher as the same level of risk, the letter highlights where dangerous biological material would have to enter the physical world.
The letter calls for safety measures for DNA production facilities to prevent them from being used with harmful materials. It also calls for safety and reliability testing for new AI models before they are released.
Openness remains part of the argument
The signatories do not argue for closing down scientific access across the board. The source says they emphasize openness and scientific freedom, while also recognizing that some AI systems may require restricted access if unresolved risks remain.
In general, the scientists support open access to AI technologies so the scientific community can study them and contribute to their development. That position reflects a balance: openness can help experts examine tools, improve them, and identify weaknesses, but access may need limits when risks are not yet addressed.
The group also committed to regularly reviewing the principles and commitments and developing new ones. That is significant because AI systems, biosafety practices, and the surrounding research environment can evolve. A static rulebook may not be enough for a changing field.
The letter also points to broad benefits through international collaboration and integrative research approaches. The implication is that safety cannot be handled only by isolated labs or individual companies. The field needs shared expectations across borders and disciplines.
How this fits into the wider AI biosecurity debate
The source also notes a recent OpenAI study on large language models and bioweapon development. According to the article, the study found that large language models such as GPT-4 only marginally facilitate the development of bioweapons because the necessary information is relatively easy to find on the Internet even without AI.
OpenAI has developed an early warning system to detect potential misuse. In the context of the scientists' open letter, that detail reinforces a broader point: AI biosecurity is not only about whether information exists, but about how systems are monitored, released, and connected to real-world production pathways.
The scientists' warning is therefore measured but serious. AI-assisted protein design could help life sciences move faster, but the same progress raises questions about biological toxins, bioweapons, DNA synthesis, model testing, and access controls.
The practical takeaway is that responsible AI in life sciences depends on more than model behavior. It also depends on procurement choices, production safeguards, ongoing evaluation, and a willingness to update rules as the technology develops.