CBRN risk
In plain wordsThe danger that AI could help someone make or use chemical, biological, radiological or nuclear weapons — for example, by explaining hard technical steps an attacker could not work out alone.
The risk that AI systems contribute to developing or using chemical, biological, radiological or nuclear weapons, usually assessed as 'uplift': how far a model increases a malicious actor's ability to cause mass casualties beyond what existing resources, such as internet search, already allow. Concerns include models giving expert-level troubleshooting for acquiring, producing or disseminating agents, especially pathogens; agents that plan or automate laboratory work; and specialized biological design tools. Developers measure it with knowledge benchmarks, expert red-teaming and uplift studiesuplift study, n.: An experiment that measures how much an AI really helps people do a task — often a dangerous one, like planning a biological attack — compared with people using only ordinary resources such as the internet., and have tied safeguards to thresholds in frameworks such as Anthropic's AI Safety LevelsAI Safety Level, n.: Anthropic's numbered scale, modeled on laboratory biosafety levels, for matching how dangerous an AI model could be with how strong the protections around it must be. and OpenAI's Preparedness FrameworkPreparedness Framework, prop. n.: OpenAI's published process for testing its AI models for catastrophic risks — such as help with biological weapons or cyberattacks — and deciding what safeguards are needed before release.; in 2025 both companies said new models might cross early biological thresholds and applied stronger protections. Mitigations include refusal training, input and output classifiersconstitutional classifiers, n.: Safety filters that watch a chatbot's inputs and outputs for dangerous requests, trained on example conversations that an AI generated from a written list of what is and is not allowed., unlearningmachine unlearning, n.: Erasing specific things a trained AI model has learned, such as dangerous lab techniques, a copyrighted novel or one person's records, far more cheaply than training the model all over again. of hazardous knowledge, structured accessstructured access, n.: Letting people use an AI system through a controlled channel, such as a website or programming interface, instead of handing over the model itself, so that the developer can limit misuse. and customer screening.
The evaluation found meaningful uplift on virology troubleshooting, so the model shipped behind ASL-3 safeguards.
Origin
'CBRN' is a defense and emergency-planning acronym that had largely replaced the older 'NBC' (nuclear, biological, chemical) by the 2000s. It entered AI policy around 2023, notably in US Executive Order 14110 (October 2023) and frontier developers' safety frameworks, where 'uplift' became the standard measure.
Related
- biosecuritybiosecurity, n.: Protecting people from the deliberate or careless spread of dangerous germs and toxins; in AI, making sure AI tools do not make it easier to create or release biological weapons.
- uplift studyuplift study, n.: An experiment that measures how much an AI really helps people do a task — often a dangerous one, like planning a biological attack — compared with people using only ordinary resources such as the internet.
- dangerous capability evaluationdangerous capability evaluation, n.: A test of whether an AI system can do things that could cause severe harm in the wrong hands or on its own — such as helping make a bioweapon, hacking computers or copying itself.
- WMDPWMDP, prop. n.: A multiple-choice test of AI knowledge that could help someone make biological, chemical or cyber weapons, built from questions near the dangerous line but not over it, and used to test ways of erasing such knowledge.
- AI Safety LevelAI Safety Level, n.: Anthropic's numbered scale, modeled on laboratory biosafety levels, for matching how dangerous an AI model could be with how strong the protections around it must be.
- dual-usedual-use, adj.: Describes technology or knowledge that can be used both to help and to harm: the same AI that finds software bugs for defenders can find them for attackers.
- misusemisuse, n.: People deliberately using AI to cause harm — for scams, harassment, propaganda, hacking or weapons — as opposed to harm from AI making mistakes or pursuing goals of its own.
Sources
- Anthropic (2025). Activating AI Safety Level 3 Protections. May 2025.
- OpenAI (2025). Preparedness Framework, version 2. April 2025.
Mentioned in
AI riskAI risk, n.: The ways AI could cause harm: people misusing it on purpose, systems behaving in ways nobody intended, and slower effects on jobs, power and society that no single person causes., misusemisuse, n.: People deliberately using AI to cause harm — for scams, harassment, propaganda, hacking or weapons — as opposed to harm from AI making mistakes or pursuing goals of its own., biosecuritybiosecurity, n.: Protecting people from the deliberate or careless spread of dangerous germs and toxins; in AI, making sure AI tools do not make it easier to create or release biological weapons., capabilitycapability, n.: Something an AI system is able to do, such as writing working code or solving math problems, and how well. Measuring capabilities is how researchers track progress and spot dangerous skills., catastrophic riskcatastrophic risk, n.: The risk that AI could cause disaster on a massive scale — many deaths or enormous damage — through weapons, cyberattacks or systems escaping human control. Several laws now define it precisely., dangerous capability evaluationdangerous capability evaluation, n.: A test of whether an AI system can do things that could cause severe harm in the wrong hands or on its own — such as helping make a bioweapon, hacking computers or copying itself., dual-usedual-use, adj.: Describes technology or knowledge that can be used both to help and to harm: the same AI that finds software bugs for defenders can find them for attackers., machine unlearningmachine unlearning, n.: Erasing specific things a trained AI model has learned, such as dangerous lab techniques, a copyrighted novel or one person's records, far more cheaply than training the model all over again., SB 1047SB 1047, prop. n.: A 2024 California bill that would have required makers of the biggest AI models to adopt safety plans and take care to prevent catastrophes. It passed the legislature, but the governor vetoed it., uplift studyuplift study, n.: An experiment that measures how much an AI really helps people do a task — often a dangerous one, like planning a biological attack — compared with people using only ordinary resources such as the internet., usage policyusage policy, n.: A company's published rules on what people may and may not use its AI for — for example, no help building weapons, no scams and no sexual content involving children., AI Safety LevelAI Safety Level, n.: Anthropic's numbered scale, modeled on laboratory biosafety levels, for matching how dangerous an AI model could be with how strong the protections around it must be., constitutional classifiersconstitutional classifiers, n.: Safety filters that watch a chatbot's inputs and outputs for dangerous requests, trained on example conversations that an AI generated from a written list of what is and is not allowed., dual-use foundation modeldual-use foundation model, n.: A US legal term for a general-purpose AI model that serves everyday uses but is capable enough that it could help someone build weapons, mount cyberattacks or evade human control., global catastrophic riskglobal catastrophic risk, n.: A risk of disaster big enough to badly damage human civilization worldwide, such as killing a large share of humanity, even if humanity eventually recovers., infohazardinfohazard, n.: Information that could cause harm simply by being shared or known — for example, details that would make a dangerous weapon easier to build., marginal riskmarginal risk, n.: How much extra danger an AI model adds compared with what people could already do without it — for instance with search engines, textbooks or models that are already freely available., Virology Capabilities TestVirology Capabilities Test, prop. n.: A hard test of practical laboratory virology know-how, used to check whether AI models know enough about working with viruses that they might help someone misuse biology., vulnerable world hypothesisvulnerable world hypothesis, n.: The idea that some future invention could make mass destruction so easy that civilization would be wrecked by default, unless it had unusually strong ways of preventing misuse., WMDPWMDP, prop. n.: A multiple-choice test of AI knowledge that could help someone make biological, chemical or cyber weapons, built from questions near the dangerous line but not over it, and used to test ways of erasing such knowledge.