Governing the AI–Bio Convergence: India’s Evolving Biosecurity Challenge

As AI–bio research moves from computational design to laboratory testing, India must address widening gaps between emerging capabilities and biosecurity safeguards through coordinated oversight across the research pipeline

Image Source: Getty Images

In August 2026, scientists at Stanford University in the United States used artificial intelligence (AI) in biological research to identify an E. coli killer and 16 new possible viruses, rekindling the conversation on AI and biological risks. The controversy over AI and bioweapons lingered for years because of the gap between possibility and evidence. Biosecurity experts voiced concerns that AI could make biological weapons easier to design, while AI researchers began implementing red-teaming and other safeguards against misuse. Meanwhile, debates about the potential consequences of AI developments remained largely hypothetical, as the capabilities required to realise these outcomes had yet to be demonstrated in a laboratory.

The new Stanford study, however, adds weight to the hypothetical. The study’s importance should be neither overstated nor underestimated. The researchers have neither created a virus that can harm humans nor shown that people without adequate biological knowledge could use AI to create pathogens. Still, the study has moved the technology from computer simulation to the real biological world. As a result, the policy question is evolving. The issue is no longer whether AI could be used to design biological agents, but how governments should respond now that this capability has been demonstrated.

What Red-Teaming Got Right and What Changed

Red-teaming exercises on AI in biosecurity, such as the one conducted by the Massachusetts Institute of Technology in 2024, did not set out to prove that AI can develop biological weapons, but to highlight the risks involved in DNA synthesis screening; however, similar risks arose at the time. While such exercises were criticised earlier, their validity increases daily. Nuances exist. The ability to produce a DNA sequence is not the same as the ability to generate a live organism. What differs now is that recent experiments have started to bridge computational and real biology. This does not mean that previous warnings are obsolete. They carry more weight now because we understand the gap between the two more accurately.

Governments need to guard against two traps: basing everything on the worst-case scenario or ignoring a danger until it materialises in a biological capability. Regulating AI models alone will not be enough. The biological workflow might include an AI model designing something, a researcher assessing it, a lab validating the options, a DNA synthesis firm producing it, and another lab investigating the resulting organism. Each of these processes can be regulated separately. Any regulating body will need to be involved in domestic scientific processes and relevant in the global dialogue.

This division creates a governance gap. AI researchers may know their models’ capabilities but not their applications. DNA synthesis firms may screen the sequences but not know their origins. Biosafety measures in labs are not necessarily connected to AI assistance.

From AI Models to Private Wet Labs

The increasing privatisation of the infrastructure that links AI to physical biology raises new governance questions. Anthropic confirmed in September 2026 that it had opened a wet biology laboratory in San Francisco, reserved in most cases for research with biological and chemical materials. The firm said the lab would be used for fundamental biological research and drug discovery and for combining AI systems with physical experimentation. Anthropic has outlined a pipeline in which its AI model, Claude, analyses biological data, forms hypotheses, and tests them physically.

What matters is both the technological and institutional nature of the change. An AI company that previously offered only computing capabilities can now conduct research directly. The boundary between the organisation creating the model and the institution carrying out biological research is blurring. Nowadays, private AI companies can build and control any part of the research pipeline themselves. The question is not whether privately held corporations ought to do biological research; it is whether public biosecurity bodies will have enough visibility and regulatory capability as biologically oriented AI is developed in private corporations.

The AI-Bio Convergence Needs New Governance

The convergence of AI and biology could offer substantial benefits, including new antibiotics, vaccines, treatments, and diagnostic tools. Research on bacteriophages was motivated by the need to develop new approaches to combating antibiotic-resistant bacteria. The appropriate response is to regulate AI-assisted biological research. Advanced systems capable of designing biological systems need to be evaluated not only on whether they prevent harmful requests but also on how they are integrated into real-world biological processes.

The convergence of AI and biology could offer substantial benefits, including new antibiotics, vaccines, treatments, and diagnostic tools.

DNA synthesis is part of the challenge. As AI develops ever more novel biological designs, simply checking for known dangerous DNA sequences may be insufficient. Collaboration among AI developers, scientists, synthesis companies, and government agencies may be required. The goal must be clear accountability as research moves from biological design to physical products. This convergence is especially pertinent to India, given its established pharmaceutical and biotechnology industries and growing capabilities in artificial intelligence and genomics.

AI-infused biology could offer considerable advantages in healthcare, agriculture, vaccine creation, and drug discovery. It also poses governance challenges because biological design can occur outside the laboratory before moving into physical experimentation.

India must create a process that involves scientists, medical professionals, security analysts, and experts on technology policy. Such a process would be necessary to address potential negative externalities and socioeconomic consequences beyond security, including those that could resemble the Bacillus thuringiensis (BT) Cotton era. This could include regulations governing high-risk artificial intelligence in biology, biological synthesis, and cooperation between AI firms, research organisations, and governmental institutions.

The next stage of AI biosecurity must account for the entire system, with regulation operating both horizontally and vertically. The Biological Weapons Convention (BWC) would remain a vital venue for discussion, but would not be sufficient on its own to ensure biological security. Governments would increasingly have to factor in the convergence of civilian AI and biotechnologies.

India’s Approach to AI-Bio Integration

To contribute to the global discourse on AI–bio integration and shape its development in India, the Department of Biotechnology (DBT) could take the following steps.

Turn Bio-AI hubs into public-private testbeds:

The Department of Biotechnology (DBT), through the Biotechnology Industry Research Assistance Council (BIRAC), may select three to five Bio-AI hubs that can formally engage frontier AI firms such as Anthropic and Google DeepMind, as well as Indian AI firms. The hubs can offer limited access to Indian biological datasets and wet labs, while the firms contribute models and technical inputs. DBT can also require biosecurity assessments for this engagement; the hubs can test any AI model deployed at Indian biological labs to determine its capabilities and limitations across different biological risks.

Create a DBT–frontier AI company “biosecurity sandbox”:

DBT can initiate a formal partnership programme with firms such as Anthropic. Instead of expecting firms to self-regulate, DBT can offer access to Indian laboratories and researchers to evaluate their models’ biological capabilities in return for permission to assess them. Anthropic claims that frontier models have been evaluated for biological risk and that it works with the government and other organisations. Anthropic also has a biology laboratory where Claude is used to develop hypotheses.

Position DBT as the institution that tests whether these safeguards work in Indian laboratories:

This would mean establishing a pipeline to oversee AI-to-DNA designs. DBT could facilitate collaboration between BIRAC-funded firms, AI design-software makers, DNA synthesisers, and biosafety panels to create a voluntary system for screening out potentially dangerous AI-generated biological designs before they are synthesised or tested. DBT could draw on the system’s outputs to craft regulations for labs and synthesis firms, and take them to international forums as part of India’s biotechnology and Biological Weapons Convention (BWC) outreach.

AI can gradually reduce the expertise and time required for biological design, while private-sector firms build the infrastructure needed to connect AI with laboratory experimentation.

Conclusion

This AI-bioweapons dilemma is evolving from the theoretical stage towards demonstrable capabilities. AI has successfully designed phage viruses for beneficial applications, without coming close to designing a deadly virus. The key is to build a bridge between computational biology and experimental biology. AI can gradually reduce the expertise and time required for biological design, while private-sector firms build the infrastructure needed to connect AI with laboratory experimentation.

Shravishtha Ajaykumar is an Associate Fellow with the Centre for Security, Strategy, and Technology at the Observer Research Foundation.

The views expressed above belong to the author(s). ORF research and analyses now available on Telegram! Click here to access our curated content — blogs, longforms and interviews.