How AI biology needs stronger safeguards

Introduction

AI could strengthen humanity’s defences against pandemics by helping scientists discover medicines, monitor emerging diseases and accelerate biological research. But the same ability to analyse and design biological systems creates a new governance challenge: advanced AI for biology is a dual-use technology, meaning the same tools that support health breakthroughs could also be misused. The central biosecurity question is therefore not whether AI biology should be slowed down, but how societies can preserve its enormous medical potential while reducing the chance of harmful applications.[National Academies]nationalacademies.orgNational Academies The Age of AI in the Life SciencesNational Academies The Age of AI in the Life Sciences

For an AI-enabled human bloom, this distinction is crucial. A future where AI helps eliminate disease and extend healthy lives depends on maintaining trust in scientific progress. If biological AI systems become easier to misuse without adequate safeguards, the technologies intended to improve human flourishing could also create new vulnerabilities. Effective biosecurity safeguards aim to keep the benefits of scientific acceleration while ensuring that powerful capabilities develop within responsible limits.[National Academies]nationalacademies.orgNational AcademiesAI Tools Can Enhance U.S. Biosecurity; Monitoring and Mitigation Will Be Needed to Protect Against MisuseMarch 14, 2025…Published: March 14, 2025

The dual-use challenge in biology

AI is becoming increasingly important in life sciences because it can process enormous biological datasets, predict molecular structures, assist with drug discovery and help researchers design new biological tools. These same capabilities create concerns because biological knowledge that is useful for medicine can sometimes also be relevant to harmful purposes. The National Academies of Sciences, Engineering, and Medicine’s 2025 consensus report on AI and life sciences describes this as a central challenge: AI-enabled biological tools may accelerate beneficial research while also introducing new biosecurity risks.[National Academies]nationalacademies.orgOpen source on nationalacademies.org.

The difficulty is that biology does not divide neatly into “safe” and “dangerous” categories. A technique developed to understand viruses, engineer therapies or improve vaccines may also increase knowledge about biological systems more generally. The goal of biosecurity is therefore not to block biological innovation, but to manage where additional caution is needed.

Several features make AI biology different from earlier scientific risks:

These challenges do not mean AI biology is uniquely dangerous. Many previous technologies, from synthetic biology to chemical manufacturing, have required safety standards because they combine valuable uses with possible misuse. The difference is that AI may accelerate the pace at which biological design becomes possible.

Testing and controlling risky capabilities

A major safeguard is testing AI systems before they are widely deployed. Traditional software testing asks whether a system works as intended; biosecurity evaluations must also ask whether a system can produce or enable harmful outcomes when deliberately misused.

This has led to the growth of red-teaming and capability evaluations, where researchers actively test whether AI models can be pushed beyond their intended boundaries. The UK’s AI Security Institute has argued that misuse safeguards need rigorous evaluation rather than assumptions that safety measures will work. Its framework recommends clearer methods for testing whether AI protections actually prevent harmful requests and actions.[AI Security Institute]aisi.gov.ukAI Security Institute Principles for evaluating misuse safeguards of frontier AI systemsAI Security Institute Principles for evaluating misuse safeguards of frontier AI systems

For biological AI, safeguards can operate at several layers:

Model design and evaluation

Developers can test advanced models against carefully designed biological risk scenarios before release. These assessments can examine whether a model provides inappropriate assistance, whether restrictions can be bypassed and whether new capabilities create risks that were not present in earlier systems.

Access controls

Not every AI capability needs to be equally available to every user or application. Systems with more advanced biological capabilities may require stronger identity checks, monitoring, institutional access or additional review processes.

Data governance

AI models depend heavily on training data. Some researchers argue that particularly sensitive biological datasets should receive greater protection, similar to other forms of sensitive information. A proposed framework for classifying dual-use pathogen data highlights the challenge of balancing scientific openness with preventing unnecessary exposure of high-risk information.[arXiv]arxiv.orgarXiv Securing Dual-Use Pathogen Data of ConcernSecuring Dual-Use Pathogen Data of ConcernFebruary 8, 2026…Published: February 8, 2026

Monitoring and accountability

Safeguards are more effective when organisations can detect misuse attempts, investigate incidents and improve systems over time. This requires logging, auditing and clear responsibility for decisions made during development and deployment.

However, safeguards are not perfect. Security researchers have repeatedly shown that AI protections can fail under unexpected conditions, including attempts to bypass restrictions. This is why biosecurity cannot depend on a single technical barrier. It requires multiple layers of defence across AI companies, research institutions, governments and international bodies.[The Guardian]theguardian.comThe Guardian AI safeguards can easily be broken, UK Safety Institute findsThe institute's research revealed that AI safeguards could be easily bypassed using basic prompts or more sophisticated jailbreaking tech…

International cooperation on AI safety

Biological risks do not respect borders, and neither do modern AI systems. A researcher, company or government in one country may develop tools whose effects are felt globally. This makes international cooperation a core part of AI biosecurity.

The challenge resembles earlier efforts to govern other dual-use technologies: societies need rules that reduce catastrophic risks without preventing legitimate scientific progress. Effective systems need cooperation on standards, information sharing, research security and responsible innovation.

The World Health Organization’s guidance on AI in health provides a broader foundation for this approach, emphasising that AI systems should be developed with accountability, transparency, equity and human benefit in mind. These principles matter for biosecurity because public trust is essential when powerful technologies influence health decisions and research priorities.[World Health Organization]who.intWorld Health OrganizationEthics and governance of artificial intelligence for healthJune 28, 2021…Published: June 28, 2021

International cooperation is developing through several routes:

The 2023 AI Safety Summit’s Bletchley Declaration reflected growing recognition among governments that advanced AI risks require international attention, including cooperation around potentially serious harms.[Reddit]reddit.comUK, US, EU and China sign declaration of AI's 'catastrophic' dangerUK, US, EU and China sign declaration of AI's 'catastrophic' dangerNovember 1, 2023…Published: November 1, 2023

Protecting innovation while preserving trust

The strongest case for AI in biology is also the reason safeguards matter. AI could contribute to faster drug discovery, improved pandemic preparedness and new medical breakthroughs that expand human health and longevity. Preventing misuse is therefore not simply a defensive goal; it is part of creating the conditions where beneficial innovation can continue.

A balanced approach avoids two extremes. One is assuming that technological progress will automatically produce good outcomes. The other is treating every powerful biological AI capability as too dangerous to develop. Both approaches ignore the central governance challenge: building institutions that allow humanity to gain the benefits while managing the risks.

For the wider AI bloom vision, biosecurity safeguards represent a form of long-term resilience. A civilisation capable of using advanced intelligence to understand biology, prevent disease and improve human wellbeing must also become capable of governing powerful tools responsibly. The future gains from AI in health depend not only on what these systems can discover, but on whether societies can ensure that their capabilities remain aligned with human flourishing.

Amazon book picks

Further Reading

Books and field guides related to How AI biology needs stronger safeguards. Use these as the next step if you want deeper reading beyond the article.

Book

The Fifth Risk

By Michael Lewis

'Will set your hair on end' Telegraph, Top 50 Books of the Year 'I forgot to breathe while reading The Fifth Risk' Michael Hofmann, TLS,...

See on Amazon

Book

The Demon in the Freezer

By Richard Preston

Jahrling is haunted by the thought that biologists in secret labs are using genetic engineering to create a new superpox virus, a smallpo...

See on Amazon

Book

The Coming Plague

By Laurie Garrett

Surveys fifty years of man's battle with communicable disease.

See on Amazon

Book

Weapons of Math Destruction

By Cathy O'Neil

'A manual for the 21st-century citizen... accessible, refreshingly critical, relevant and urgent' - Financial Times 'Fascinating and deep...

See on Amazon

Browse more on Amazon:The Fifth RiskThe Demon in the FreezerThe Coming Plague

As an Amazon Associate I earn from qualifying purchases.

eBay marketplace picks

Marketplace Samples

Live-tested eBay searches with available results related to this page.

Shop location

UsingUSA USA USAUKCanadaAustraliaIreland

Selected frombiology model oneBay.co.uk.

Current eBay listing

The Great Mental Models Volume 2 - Physics, Chemistry And Biology

GBP 26.70 | Shipping GBP 3.38Like NewSeller: geminits (100.0% positive)Ships from: GB

View listing on eBay

Current eBay listing

Edu-Science Human Torso Model Kit Anatomy Biology with Removable Organs

GBP 69.00 | Free shippingUsedSeller: zamzamlab (98.9% positive)Ships from: GB

View listing on eBay

Current eBay listing

Human Chromosome Model Practical Easy Assembly Easy to Use Biology Model Set

GBP 12.05 | Free shippingNewSeller: gorgeousbritishloft (99.5% positive)Ships from: CN

View listing on eBay

Browse more oneBay.co.uk

Example items shown for inspiration; availability and pricing can change. Branchoria may earn a commission if you purchase through outbound eBay links.

Endnotes

1. Source: arxiv.org
Title: arXiv Securing Dual-Use Pathogen Data of Concern
Link:https://arxiv.org/abs/2602.08061

Published: February 8, 2026

2. Source: reddit.com
Title: UK, US, EU and China sign declaration of AI’s ‘catastrophic’ danger
Link:https://www.reddit.com/r/artificial/comments/17lkvz6

Published: November 1, 2023

3. Source: nationalacademies.org
Title: National Academies The Age of AI in the Life Sciences
Link:https://www.nationalacademies.org/projects/DELS-BLS-24-04/publication/28868

4. Source: nationalacademies.org
Link:https://www.nationalacademies.org/news/ai-tools-can-enhance-u-s-biosecurity-monitoring-and-mitigation-will-be-needed-to-protect-against-misuse

Published: March 14, 2025

5. Source: nationalacademies.org
Link:https://www.nationalacademies.org/publications/28868?page=57

6. Source: aisi.gov.uk
Link:https://www.aisi.gov.uk/research/principles-for-evaluating-misuse-safeguards-of-[frontier

7. Source: theguardian.com
Title: The Guardian AI safeguards can easily be broken, UK Safety Institute finds
Link:https://www.theguardian.com/technology/2024/feb/09/ai-safeguards-can-easily-be-broken-uk-safety-institute-finds

8. Source: who.int
Link:https://www.who.int/publications/i/item/9789240029200

Published: June 28, 2021

9. Source: aiwiki.ai
Title: Bletchley Declaration | AI Wiki
Link:https://aiwiki.ai/wiki/bletchley_declaration

Published: June 21, 2026

10. Source: GOV.UK
Title: www.gov.uk Data and AI Ethics Framework
Link:https://www.gov.uk/government/publications/data-ethics-framework/data-and-ai-ethics-framework

11. Source: who.int
Link:https://www.who.int/publications/i/item/9789240108868

12. Source: GOV.UK
Link:https://www.gov.uk/government/publications/ai-safety-summit-2023-the-bletchley-declaration/the-bletchley-declaration-by-countries-attending-the-ai-safety-summit-1-2-november-2023
Published: november 2023

13. Source: uwnxt.nationalacademies.org
Link:https://uwnxt.nationalacademies.org/read/28868

14. Source: uwnxt.nationalacademies.org
Link:https://uwnxt.nationalacademies.org/read/28868/chapter/2

15. Source: uwnxt.nationalacademies.org
Link:https://uwnxt.nationalacademies.org/read/28868/chapter/3

16. Source: uwnxt.nationalacademies.org
Link:https://uwnxt.nationalacademies.org/projects/DELS-BLS-24-04/event/43297

17. Source: GOV.UK
Title: www.gov.uk A I Safety Institute approach to evaluations
Link:https://www.gov.uk/government/publications/ai-safety-institute-approach-to-evaluations/ai-safety-institute-approach-to-evaluations

18. Source: GOV.UK
Title: ai safety summit 2023 the bletchley declaration
Link:https://www.gov.uk/government/publications/ai-safety-summit-2023-the-bletchley-declaration

19. Source: GOV.UK
Link:https://www.gov.uk/government/news/countries-agree-to-safe-and-responsible-development-of-frontier-ai-in-landmark-bletchley-declaration?bxid=5cec23dffc942d3ada01da49&cndid=38838028&esrc=Wired_etl_load

20. Source: GOV.UK
Title: ai safety summit 2023 the bletchley declaration.es 419
Link:https://www.gov.uk/government/publications/ai-safety-summit-2023-the-bletchley-declaration.es-419

21. Source: industry.gov.au
Title: bletchley declaration countries attending ai safety summit 1 2 november 2023
Link:https://www.industry.gov.au/publications/bletchley-declaration-countries-attending-ai-safety-summit-1-2-november-2023
Published: november 2023

22. Source: who.int
Link:https://www.who.int/publications-detail-redirect/9789240037403

23. Source: who.int
Link:https://www.who.int/news/item/28-06-2021-who-issues-first-global-report-on-ai-in-health-and-six-guiding-principles-for-its-design-and-use

24. Source: who.int
Title: Health Ethics & Governance
Link:https://www.who.int/teams/health-ethics-governance/emerging-technologies/big-data-and-artificial-intelligence

25. Source: who.int
Link:https://www.who.int/publications/b/81594

26. Source: who.int
Link:https://www.who.int/publications/b/58847

27. Source: who.int
Link:https://www.who.int/teams/health-ethics-governance/emerging-technologies

28. Source: who.int
Link:https://www.who.int/groups/technical-advisory-group-on-the-responsible-use-of-the-life-sciences-and-dual-use-research-%28tag-ruls-dur%29

29. Source: emro.who.int
Title: int WH O EMRO
Link:https://www.emro.who.int/emhj-volume-30-2024/volume-30-issue-6/adhering-to-ethics-guidelines-in-biomedical-research-and-medical-practice-is-crucial-to-save-lives.html

30. Source: nationalacademies.org
Link:https://www.nationalacademies.org/read/28868

31. Source: nationalacademies.org
Link:https://www.nationalacademies.org/publications/28868?page=552&publication_date%5Bquick%5D=past_5_years&sort=asc

32. Source: nap.nationalacademies.org
Link:https://nap.nationalacademies.org/collection/69/biosecurity

33. Source: nap.nationalacademies.org
Link:https://nap.nationalacademies.org/catalog/27971/strategic-report-on-research-and-development-in-biotechnology-for-defense-innovation?s=z1120

34. Source: aisi.gov.uk
Link:https://www.aisi.gov.uk/blog/principles-for-safeguard-evaluation

35. Source: aisi.gov.uk
Link:https://www.aisi.gov.uk/frontier-ai-trends-report

36. Source: aisi.gov.uk
Link:https://www.aisi.gov.uk/research

37. Source: aisi.gov.uk
Link:https://www.aisi.gov.uk/research-agenda

38. Source: aisi.gov.uk
Link:https://www.aisi.gov.uk/blog

39. Source: oecd.org
Link:https://www.oecd.org/en/topics/sub-issues/synthetic-biology.html

40. Source: regulations.ai
Title: Bletchley Declaration
Link:https://regulations.ai/glossary/bletchley-declaration

Additional References

41. Source: youtube.com
Link:https://www.youtube.com/watch?v=SvPcxRiFauY

42. Source: youtube.com
Link:https://www.youtube.com/watch?v=S0idP71jyGg

43. Source: youtube.com
Title: How can the BWC address the dual-use challenges of AI-driven biodesign tools?
Link:https://www.youtube.com/watch?v=lBgmASzueR0

44. Source: youtube.com
Title: From AI Boom To Bioweapons Battle: CEOs Demand New Safeguards
Link:https://www.youtube.com/watch?v=MZAHa4G–7s

45. Source: youtube.com
Title: In New Delhi, Hassabis Warns of AI Biosecurity and Cybersecurity Risks
Link:https://www.youtube.com/watch?v=_au-zciGxD8

46. Source: oecd.org
Link:https://www.oecd.org/en/publications/anticipatory-governance-for-responsible-innovation-in-synthetic-biology_864dd003-en/full-report/component-5.html

47. Source: oecd.org
Link:https://www.oecd.org/en/publications/anticipatory-governance-for-responsible-innovation-in-synthetic-biology_864dd003-en/full-report/component-4.html

48. Source: nature.com
Link:https://www.nature.com/articles/s41587-025-02650-8

49. Source: oecd.ai
Link:https://oecd.ai/en/ai-publications/advancing-accountability

50. Source: doi.org
Link:https://doi.org/10.1007/s43681-025-00872-9