Examples
Examples from published documents
- Google DeepMind builds threat actor types from "group size, resource availability, expertise, and supply chain access"; one example is "a lone wolf actor, with fund availability <$100,000, undergraduate studies in chemistry, and limited supply chain access via friends or colleagues" (Gemini 3.7 Flash FSF report, p.8)
- Anthropic's novel chemical and biological weapons threshold is calibrated to a "well-resourced team", footnoted as "a program with millions of dollars available and support from recent PhD's from top-tier international programs" (Anthropic Responsible Scaling Policy v3.4, section 1 table, footnote 5)
- Meta's chemical and biological outcomes are split by actor: CB 1 concerns weapons "for low and moderate skill actors" and CB 2 concerns agents "for high-skilled actors" (Meta Advanced AI Scaling Framework v2, section 3.4.2)
Counter-examples
Not to be confused with
- The threat actors named in a security goal are a different construct: the EU Code of Practice's Security Goal and xAI's framework list the attackers that weight-security measures must withstand, not the misusers a capability threshold is calibrated against.
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.
- Threat Actor Capability (TAC) levels
- Threat Actor Capability (TAC) levels are Google DeepMind's four-step cyber attacker scale, from TAC-1, a "programmer with limited cybersecurity knowledge", to TAC-4, a "95th+ percentile cybersecurity expert (e.g., nation-state-sponsored groups)".
- Source: Gemini 3.7 Flash FSF report
- Non-state external threat
- A non-state external threat, in the EU Code of Practice, is a hostile operation by non-state actors roughly comparable to ten experienced cybersecurity professionals, spending several months and up to EUR 1 million, without pre-existing access to the target.
- Source: EU GPAI Code of Practice, Safety and Security Chapter
- Sophisticated insider
- A sophisticated insider, in Anthropic's August 2026 Risk Report, is one of the "insiders who have persistent access or can request time-limited access to systems that process model weights". Anthropic places this group outside its ASL-3 security standard.
- Source: Anthropic Risk Report, August 2026
Source of record
Where this definition comes from
Anthropic Risk Report, August 2026, §4.1, §2.11.1, §6.4.1 fn90
“CB-1: "Individuals or small groups with limited resources"; CB-2: "Moderately resourced threat actors (including, for example, expert-backed teams)" (§4.1). "Sophisticated insiders": "insiders who have persistent access or can request time-limited access to systems that process model weights" (§6.4.1 fn 90).”
https://www-cdn.anthropic.com/f61d49fa5596956a5dec75fea0e973bf6a6a8378/Redacted%20Risk%20Report%20August%202026%20.pdfGemini 3.7 Flash FSF report / FSF v3.1, report p.16
“Threat Actor Capability levels "TAC-1" ("programmer with limited cybersecurity knowledge (e.g., 'script kiddie')") to "TAC-4" ("95th+ percentile cybersecurity expert (e.g., nation-state-sponsored groups)") (report p.16).”
https://storage.googleapis.com/deepmind-media/gemini/gemini_3-7_flash_fsf_report.pdf
Crosswalk
How named organisations use this concept
| Organisation | Their term, as published | Match & verification | Source |
|---|---|---|---|
| AnthropicShadow mapping Anthropic Risk Report, August 2026 | “CB-1: "Individuals or small groups with limited resources"; CB-2: "Moderately resourced threat actors (including, for example, expert-backed teams)" (§4.1). "Medium-strength attacker": "an attacker with the strength that a human red-teamer shows when editing a transcript without iterating against a monitor ..." (§2.11.1). "Sophisticated insiders": "insiders who have persistent access or can request time-limited access to systems that process model weights" (§6.4.1 fn 90).” | exactEQ confidence: high | Anthropic Risk Report, August 2026 |
| OpenAIShadow mapping Preparedness Framework v2 / GPT-5.6 deployment safety report | “High (Biological and Chemical): "The model can provide meaningful counterfactual assistance (relative to unlimited access to baseline of tools available in 2021) to “novice” actors (anyone with a basic relevant technical background) that enables them to create known biological or chemical threats." Critical: "The model can enable an expert to develop a highly dangerous novel threat vector" (Table 1). GPT-5.6: "Spray and pray operations": "broad exploitation by moderately skilled, low-resourced individuals and small groups" (s.9.4.1.2).” | closeCL confidence: high | OpenAI Preparedness Framework v2 |
| Google DeepMindShadow mapping Gemini 3.7 Flash FSF report / FSF v3.1 | “"Provides low to medium resourced actors uplift in reference scenarios" (CBRN Uplift 1 CCL). Threat Actor Capability levels "TAC-1" ("programmer with limited cybersecurity knowledge (e.g., 'script kiddie')") to "TAC-4" ("95th+ percentile cybersecurity expert (e.g., nation-state-sponsored groups)") (report p.16). FSF refers to RAND "OC3 groups" without defining them (p.11).” | exactEQ confidence: high | Gemini 3.7 Flash FSF report |
| xAIShadow mapping xAI Frontier AI Framework (30 Jun 2026) | “Security Goal threat actors: "sophisticated non-state actors, insider threats, state-sponsored actors and other foreseeable actors" (s.2.4). This is a security-only construct, narrower than the malicious-use threat-actor profiles used elsewhere in the corpus.” PDF metadata /Title reads "Privileged/Confidential DRAFT working FRAMEWORK DOC"; no xAI statement disambiguating draft vs. final was found. | narrowNR confidence: medium | xAI Frontier AI Framework (30 Jun 2026, draft-marked) |
| MetaShadow mapping Meta Advanced AI Scaling Framework v2 | “Scenarios cover "both state and non-state actors" and distinguish "high- and low-skill actors" (para., §1.1); CB 1: "Proliferation of known medium-impact biological or chemical weapons for low and moderate skill actors" (§3.4).” | closeCL confidence: high | Meta Advanced AI Scaling Framework v2 |
| EUShadow mapping EU GPAI Code of Practice, Safety and Security chapter | “Measure 6.1 "Security Goal": Signatories "will define a goal that specifies the threat actors that their security mitigations are intended to protect against ..., including non-state external threats, insider threats, and other expected threat actors, taking into account at least the current and expected capabilities of their models" — a security-only threat-actor construct, distinct from malicious-use threat-actor profiles used elsewhere in the corpus, confirming xAI's own adoption of the same term.” Row also cites xAI's Frontier AI Framework, whose PDF metadata is marked "DRAFT working FRAMEWORK DOC" with no disambiguating xAI statement found. | closeCL confidence: high | EU GPAI Code of Practice, Safety and Security chapter |
| MicrosoftShadow mapping Microsoft Frontier Governance Framework (Feb 2026) | “CBRN High: "The model provides a meaningful uplift to an expert's (PhD level education in related fields) ability to develop and deliver a novel CBRN threat." "The model provides a meaningful uplift to a medium-skilled actor's (e.g., STEM education) ability to develop and deliver a known CBRN threat." Critical: "significantly lowers the barriers to a low-skilled actor developing and delivering a known CBRN threat" (Appendix I, p.11).” | closeCL confidence: high | Microsoft Frontier Governance Framework (Feb 2026) |
| AmazonShadow mapping Amazon Frontier Model Safety Framework | “CBRN threshold: "would enable a non-subject matter expert to reliably produce and deploy a CBRN weapon"; Offensive Cyber Operations threshold: "a moderately skilled actor (e.g., an individual with undergraduate level understanding of offensive cyber activities or operations)" (s.1, p.2).” Actor labels are not defined further in the framework. | closeCL confidence: high | Amazon Frontier Model Safety Framework |
| AnthropicShadow mapping Anthropic Responsible Scaling Policy v3.4 | “Non-novel CB threshold: "individuals or groups with basic technical backgrounds (e.g., undergraduate STEM degrees)"; novel CB threshold: "a well-resourced team", footnoted as "a program with millions of dollars available and support from recent PhD's from top-tier international programs" (section 1 table, pp.6-7).” | closeCL confidence: high | Anthropic Responsible Scaling Policy v3.4 |
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
Related, not mapped
Pointers that are not crosswalk claims
These sources mention this concept but do not define or map it clearly enough to count as a crosswalk row — noted here so the research is visible without overstating it as a mapping.
- Frontier Model Forum
"Heightened Marginal Risk" examples (e.g. "enabling low-skilled actors to build chemical weapons previously restricted to well-resourced states") and FMF35's "Threats" examples that mention "potential threat actors" are related context, not a defined threat-actor-profile construct — related, not a mapping.
FMF Risk Taxonomy and Thresholds / Information Sharing, Incident Reporting and Incident Response







