Examples
Examples from published documents
- Google DeepMind's Frontier Safety Framework uses four risk domains: CBRN, Cyber, Harmful Manipulation, and a combined "ML R&D and Misalignment" domain (Google DeepMind Frontier Safety Framework v3.1, glossary, p.18)
- Meta's framework covers catastrophic risks in three areas: "Chemical & Biological, Cybersecurity, and Loss of Control" (Meta Advanced AI Scaling Framework v2, preamble, p.2)
- The EU Code lists four specified systemic risks: CBRN, loss of control, cyber offence and harmful manipulation (EU GPAI Code of Practice, Safety and Security Chapter, Appendix 1.4)
Counter-examples
Not to be confused with
- A threat model is a specific causal route to harm inside a domain, such as a novice actor using a model to produce a biological weapon; the domain is the broader category.
- A risk level or tier rates how much risk a model poses within a domain; it is not the domain itself.
Related terms
Related AI safety terms
Terms defined under this element that are not NIKOLAI elements themselves, each as the cited document uses it. Every term is also listed in the AI safety glossary.
- CBRN
- CBRN stands for chemical, biological, radiological and nuclear. As a risk domain it covers AI enabling CBRN attacks or accidents, for example by lowering barriers to entry for malicious actors in designing or acquiring such weapons.
- Source: EU GPAI Code of Practice, Safety and Security Chapter
- Systemic risk
- Systemic risk, under EU AI Act Article 3(65), is a risk specific to the high-impact capabilities of general-purpose AI models that can have a significant impact on the EU market and be propagated at scale across the value chain.
- Source: EU AI Act Article 3 (Definitions)
Source of record
Where this definition comes from
EU GPAI Code of Practice, Safety and Security Chapter, Appendix 1.4
“Appendix 1.4 "Specified systemic risks": "(1) Chemical, biological, radiological and nuclear ... (2) Loss of control ... (3) Cyber offence ... (4) Harmful manipulation: Risks from enabling the strategic distortion of human behaviour or beliefs by targeting large populations or high-stakes decision-makers through persuasion, deception, or personalised targeting ..." -- the taxonomy xAI's FAIF footnotes say it adopts.”
https://ec.europa.eu/newsroom/dae/redirection/document/118119Anthropic Advanced AI Framework, p.4
“"Enumerated Risk categories": "catastrophic risks related to (a) biological weapons, (b) offensive cyber operations, (c) loss of control, and (d) automated research and development in key domains that could accelerate or amplify risks (a)-(c)."”
https://www-cdn.anthropic.com/files/4zrzovbb/website/0a58d567024a8b448ff15158ebc3625328dfcc1f.pdfFrontier Model Forum, Risk Taxonomy and Thresholds technical report, s2.4, p.9
“"Chemical, Biological, Radiological, and Nuclear (CBRN) Threats", "Advanced Cyber Threats", "Advanced Autonomous Behavior Threats"”
https://www.frontiermodelforum.org/technical-reports/risk-taxonomy-and-thresholds/
Crosswalk
How named organisations use this concept
| Organisation | Their term, as published | Match & verification | Source |
|---|---|---|---|
| AnthropicShadow mapping Risk Report / Advanced AI Framework | “Four threat models in three sections: misalignment in high-stakes settings (Autonomy TM1), automated R&D in key domains (TM2), CB-1 and CB-2 (para.). AAF "Enumerated Risk categories": "catastrophic risks related to (a) biological weapons, (b) offensive cyber operations, (c) loss of control, and (d) automated research and development in key domains that could accelerate or amplify risks (a)-(c)." (p.4)” | exactEQ confidence: high | Anthropic Advanced AI Framework, legislative proposal |
| OpenAIShadow mapping Preparedness Framework v2 / Frontier Governance Framework | “PF "Tracked Categories": Biological and Chemical; Cybersecurity; AI Self-improvement. "Research Categories": Long-range Autonomy; Sandbagging; Autonomous Replication and Adaptation; Undermining Safeguards; Nuclear and Radiological (Tables 1 and 2). FGF risk categories: "Cyber offense", "CBRN", "Harmful manipulation", "Loss of control" (§2.1).” | exactEQ confidence: high | OpenAI Preparedness Framework v2 |
| Google DeepMindShadow mapping Frontier Safety Framework v3.1 | “"CBRN", "Cyber", "Harmful Manipulation", "ML R&D and Misalignment" (glossary, p.18).” | exactEQ confidence: high | Google DeepMind Frontier Safety Framework v3.1 |
| xAIShadow mapping Frontier AI Framework, 30 Jun 2026 | “"CBRN Risks", "Offensive Cybersecurity Risks", "Loss of Control Risks", "Harmful Manipulation Risks" (s.1), with fn 1: "Terminology used in the The Safety and Security Chapter of the General-Purpose AI Code of Practice developed under the EU AI Act." Harmful Manipulation notably has no "Addressing" subsection unlike the other three domains.” This source's PDF metadata /Title reads "Privileged/Confidential DRAFT working FRAMEWORK DOC"; no xAI statement disambiguating draft vs. final status was found, so its status remains unresolved. The EQ rating covers the domain naming only; the framework has no "Addressing" subsection for Harmful Manipulation, which is a declared-but-undefined gap within this row. | exactEQ confidence: high | xAI Frontier AI Framework, 30 Jun 2026 |
| MetaShadow mapping Advanced AI Scaling Framework v2 | “"Chemical & Biological, Cybersecurity, and Loss of Control" (preamble p.2). "Risk domain: is used to describe the thematic grouping that a set of catastrophic outcomes belong to" (Appendix I).” | exactEQ confidence: high | Meta Advanced AI Scaling Framework v2 |
| EUShadow mapping Safety and Security chapter | “Appendix 1.4 "Specified systemic risks", verbatim: "(1) Chemical, biological, radiological and nuclear: Risks from enabling chemical, biological, radiological, and nuclear (CBRN) attacks or accidents ... (2) Loss of control: Risks from humans losing the ability to reliably direct, modify, or shut down a model ... (3) Cyber offence: Risks from enabling large-scale sophisticated cyber-attacks, including on critical systems ... (4) Harmful manipulation: Risks from enabling the strategic distortion of human behaviour or beliefs by targeting large populations or high-stakes decision-makers through persuasion, deception, or personalised targeting ..." -- the taxonomy xAI's FAIF footnotes say it adopts.” | exactEQ confidence: high | EU GPAI Code of Practice, Safety and Security Chapter |
| California SB 53Shadow mapping SB 53 statute text | “Catastrophic-risk pathways: (A) CBRN "expert-level assistance"; (B) conduct "with no meaningful human oversight ... that is either a cyberattack or ... would constitute the crime of murder, assault, extortion, or theft"; (C) "Evading the control of its frontier developer or user" (22757.11(c)).” SB 53 frames these as incident *pathways* (mechanisms), not a domain taxonomy per se; mapped here as close because the pathways cluster along the same CBRN / cyber / loss-of-control lines as the other sources' domains. | closeCL confidence: medium | California SB 53 |
| US Government (Executive Order 14409 / NIST CAISI)Shadow mapping Executive Order 14409 / CAISI bulletin | “EO 14409 addresses only "advanced cyber capabilities" (Sec. 3(a)); CAISSI (formerly CAISI) frames its agreements as "Frontier AI National Security Testing".” | narrowNR confidence: medium | Executive Order 14409 |
| Frontier Model ForumShadow mapping Risk Taxonomy and Thresholds technical report | “"Chemical, Biological, Radiological, and Nuclear (CBRN) Threats", "Advanced Cyber Threats", "Advanced Autonomous Behavior Threats" (s2.4, p.9).” | exactEQ confidence: high | Frontier Model Forum, Risk Taxonomy and Thresholds technical report |
What do these codes mean?
- exact
- The source term is equivalent to this element
- close
- The source term is close but not equivalent to this element
- broad
- The source term is broader than this element
- narrow
- The source term is narrower than this element
- none
- No mapping claim — used for false-friend and declared-but-undefined rows
- EQ
- Equivalent
- CL
- Close
- BR
- Source is broader than the element
- NR
- Source is narrower than the element
- FF
- False friend — same or similar label, different meaning
- DU
- Declared but undefined by the source
- UV
- Unverified
Divergence
Where sources materially disagree
The domain lists overlap but do not line up. AI R&D is an OpenAI Tracked Category, part of Google DeepMind's merged "ML R&D and Misalignment" domain, a threat model in Anthropic's Risk Report, and an enabling capability under Meta's Loss of Control domain. Radiological and nuclear risks are a Research Category for OpenAI and an "emerging" area for Meta, but part of CBRN for the EU Code and Google DeepMind. "Systemic risk" also means different things: in the EU AI Act it is a risk specific to high-impact general-purpose models, while the International AI Safety Report uses it for risks from widespread deployment across society and notes that the EU Act uses the term differently. Risk-domain stays in N1 because it fixes a report's scope; N2's threat-model and risk-pathway elements describe causal routes within a domain.
Gap
"Harmful manipulation" is a named domain in OpenAI's Frontier Governance Framework, Google DeepMind's Frontier Safety Framework, xAI's Frontier AI Framework and the EU Code (Appendix 1.4, verified), but OpenAI calls its approach exploratory and publishes no tiers for it, and xAI's framework gives it no "Addressing" subsection. The research did not find a quantified threshold for this domain in any lab framework.







