biosecurity

advancedspreading

The totality of rules and protective measures designed to ensure that no one can intentionally cause harm using dangerous pathogens, even with the help of artificial intelligence.

In plain terms

Biosecurity originally referred to laboratories, epidemiology, and animal health: it concerns ensuring that dangerous viruses, bacteria, and sensitive knowledge about them do not fall into the wrong hands. In the AI era, it has acquired a new meaning: it examines whether a language model or an agent could help someone create a biological weapon and how this can be prevented.

Think of a pharmacy. The pharmacist is happy to explain how a drug works, but certain medicines are only dispensed with a prescription, and the most dangerous ones are kept in a safe. AI developers expect the same from their models: teach biology, support research, but do not provide steps leading to weapons of mass destruction.

It has moved to the forefront of AI safety because the harm here can be irreversible and mass-scale. Bad advice on a food recipe is an annoyance; on a pathogen that causes an epidemic, it is a catastrophe. That is why companies and regulators rank this area among the most serious risks.

How it works

Survey: before release, developers test with experts how much further a layperson or a beginner researcher can get with the model compared to without it (this is called capability uplift).
Adversarial testing: biologists and security experts deliberately attempt to prompt the model into giving dangerous instructions to reveal its weaknesses.
Defense layers: the model is trained to reject dangerous requests, while separate filters monitor incoming queries and outgoing responses.
Thresholds and decisions: if a model's capabilities exceed a preset danger level, stricter safeguards are required, or the release is postponed.
Protection in the physical world: many DNA manufacturing companies screen orders (this is voluntary in most countries, not mandatory), laboratories working with dangerous pathogens require permits, and international agreements ban biological weapons.

An everyday example

You are preparing for a biology class and ask the chatbot how viruses reproduce: you get a detailed, understandable answer. If, however, you ask how a pathogen could be made more infectious, the system politely refuses. That is the biosecurity safeguard at work.

Who is affected

everyoneA deliberately triggered epidemic does not stop at the national border, so its prevention is in everyone's interest.

patient, doctorThe same AI that helps develop medicines and vaccines can also be misused, which is why medical research is subject to strict limitations. → Health

developerIf you build a model or agent, you must account for the fact that biology-related requests are handled by special rules and filters. → Code

parent, student, teacherDuring training, a chatbot may reject an innocent biological or chemical question as a precaution. → Education & Family

A common misconception

MythWith AI, anyone can now assemble a biological weapon at home.

In realityKnowledge is only one obstacle: laboratory equipment, raw materials, and much practical experience are also needed. However, the risk increases as models develop, so defenses are being built now.

Limits and risks

Filters are not perfect: they can sometimes be bypassed with clever phrasing, while at other times they reject harmless school questions.
It is difficult to measure exactly how much a model actually helps a bad actor, which is why experts' risk assessments also differ.
Protection can be removed from freely downloadable open models, so the developer's rules count for little there.
Overly strict restrictions can hinder well-meaning researchers, doctors, and teachers.

Similar, but not the same

laboratory biosafety – It protects laboratory workers and the environment against accidental incidents and infections, whereas biosecurity prevents intentional misuse; in Hungarian, both are often translated as 'biological safety,' which can cause confusion.
cybersecurity – It protects computer systems and data from attacks, while biological safety prevents harm caused by living pathogens.
AI safety – AI safety covers all AI risks; biosecurity is just one, particularly severe subfield.

Related concepts

Broader

AI safety, catastrophic risk, risk management

Narrower

capability uplift, DNA synthesis screening, assessment of dangerous capabilities

On the edge

dual-use technology, red teaming, content filtering, cybersecurity, Responsible scaling policy

A brief history

The international treaty banning biological weapons was opened for signature in 1972 and entered into force in 1975. The term "biosecurity" in the sense of preventing intentional misuse spread later, particularly after the 2001 anthrax letter attacks in the US. In the context of AI, it came into focus in 2023: the closing declaration of the AI Safety Summit held at Bletchley Park in November specifically mentioned biotechnological risks.

Where things stand

By the end of 2026, major AI developers regularly test their models for biological risks before release. The EU AI Act requires providers of general-purpose AI models posing systemic risks to assess and mitigate such risks; these obligations have applied since August 2025. However, there is still no unified, mandatory international standard.

Status: 2026 November

In the news

Appears in 2 Agent Age stories; the curve shows weekly occurrence in the news magazine.

UK sets aside about $800 million for AI security, including a biosecurity labOct 23, 2026 · Agent Age
Stanford researchers design 16 new viruses with AI, raising biosecurity concernsAug 6, 2026 · Health

Also known as

biosecurity · biosecurity · biological defense · biosecurity · bio-security · AI biosecurity · biorisk · CBRN risk

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.