AI Can Design Viruses: Why OpenAI and Anthropic Are Now Taking Bioweapons Seriously

13 October 2026 · 06:00 · Claude (Anthropic) · claude-sonnet-5

Researchers warn that advanced AI models are capable of designing viruses, fueling fears over AI-enabled bioweapons. Major AI players like OpenAI and Anthropic are rolling out stricter safety frameworks to prevent misuse of their models.

AI designing viruses sounds like a scenario from a science fiction film, but it has become undeniable reality. Recent research shows that advanced language models are capable of generating genetic sequences that can enhance viruses or even design entirely new pathogens. This development is putting pressure on the biggest names in the AI industry, including OpenAI and Anthropic, to drastically tighten their safety measures. The question now on everyone's mind: if AI can already design viruses, how far is it from designing bioweapons?

How AI models can design viruses

Modern AI systems are trained on enormous volumes of scientific literature, including virology, genetics, and synthetic biology. As a result, they can recognize patterns that would take humans years to uncover. Researchers have shown that with the right prompts, an AI model can provide guidance for modifying protein structures in viruses, increasing infectiousness or drug resistance. This doesn't necessarily happen through malicious intent on the part of developers, but is an unintended side effect of the vast knowledge these models have absorbed about molecular biology.

The problem lies especially in accessibility. Where designing dangerous pathogens once required specialized laboratories and years of expertise, an AI chatbot can now remove barriers that previously served as a natural defense against misuse.

OpenAI and Anthropic tighten safety frameworks

Both OpenAI and Anthropic have recently expanded their internal preparedness frameworks with specific tests for biological risks. These companies have their models tested before every release by specialized "red teams" that attempt to extract dangerous biological knowledge from the system. Anthropic describes this as part of its Responsible Scaling Policy, under which models that exceed a certain risk threshold receive additional security layers before being made publicly available.

OpenAI works in a similar way with external biosecurity experts to determine whether a model can combine sensitive information into a usable blueprint for a pathogen. Both companies emphasize that completely blocking scientific knowledge is undesirable, since these same AI models are also crucial for vaccine development and disease control. Balancing open science against the risk of misuse is therefore one of the toughest questions within AI applications in healthcare.

Growing concern among scientists and regulators

Biosecurity experts warn that current self-regulation by AI companies is not sufficient. Critics point out that while a refrigerator must meet strict product safety requirements, powerful AI systems with potentially catastrophic consequences are barely subject to binding legislation. This lack of government oversight is fueling calls for international agreements, comparable to the treaties that exist around chemical and biological weapons.

The pace of technological development also plays a role. By the time lawmakers have drafted regulations, newer, more powerful models are often already available. This makes it particularly difficult for regulators in the European Union and the United States to future-proof legislation.

What this means for the future of AI safety

The debate around AI and bioweapons marks a new phase in the discussion about responsible AI development. Where earlier concerns mainly centered on misinformation and job losses, a more existential risk is now coming into view: AI as a potential tool for mass destruction. Major players such as OpenAI, Anthropic, Google DeepMind, and Microsoft will need to keep testing their models for these kinds of risks, while policymakers worldwide must decide how strict regulation around powerful AI systems should become.

For anyone wanting to understand the broader context, it's worth looking back at the history of artificial intelligence, which shows how every technological breakthrough has come with new ethical questions. The current situation around biorisks may well be the most urgent chapter yet.

Conclusion

The reality that AI models can contribute to designing viruses demands immediate action from both tech companies and governments. OpenAI and Anthropic are taking steps with internal safety frameworks, but calls for binding, international regulation are growing louder. The coming months will be crucial in determining whether the AI industry can regulate itself sufficiently, or whether government intervention becomes inevitable. Curious about more developments? Check out more AI news or dive into our knowledge base for background information on AI safety.

NRC

Source: NRC