LowCBRN Uplift●Realized

CBRN / dangerous capability uplift

Description

The model provides meaningful uplift toward chemical, biological, radiological, or nuclear or other mass-harm capabilities.

Example scenario

A misused internal tool is probed for synthesis instructions; controls must hard-refuse.

Real-world evidence●Realized

Italy's Garante blocked ChatGPT in March 2023 citing GDPR violations related to unlawful personal data processing, and OpenAI subsequently faced formal enforcement proceedings across multiple EU jurisdictions. These are confirmed, publicly documented regulatory enforcement actions against AI systems processing personal data in violation of data protection law.

Italy blocks ChatGPT over GDPR personal data concernsGarante per la Protezione dei Dati Personali · 2023 CNIL fines Clearview AI for unlawful facial recognition data processingCNIL (French data protection authority) · 2022

Primary mitigations

Hazardous-capability classifiers
hard refusals
uplift evals
access gating
monitoring.

Detection signals

CBRN-benchmark refusal rate; dangerous-query monitoring.

Mitigating controls

Non-agentic controls

ZYC-SEC-001Non-Agentic

Adversarial Testing

Security

ZYC-ETH-001Non-Agentic

Non-Maleficence

Ethical Governance

ZYC-ACCU-001Non-Agentic

Lexical Similarity

Accuracy

ZYC-LEGAL-001Non-Agentic

AI Regulation Compliance

Legal Compliance

Risk scoring

Severity

High

score 3

Likelihood

Low

score 1

Risk

Low

Control depth

4 mapped (below avg 5.2)

Framework refs

5/8 taxonomies

Framework mappings

Regulatory4/5

EU AI Act: Art. 10(2)(f) (data governance — training data collection must comply with applicable data protection law including lawful basis); Art. 10(5) (special conditions for processing sensitive personal data in high-risk AI systems for bias monitoring — must comply with GDPR Art. 9); Art. 13 (transparency — information provided to deployers must include data processing scope). GDPR remains the lex specialis; the EU AI Act is explicitly complementary (Recital 9) and does not override or replace GDPR obligations.
DORA: — Not covered by DORA. GDPR and data protection regulatory compliance are outside DORA's scope. DORA Art. 9(2) addresses data confidentiality as an ICT security property but does not impose or enforce data protection law compliance.
RBI: Safety, Resilience and Sustainability
NIST AI RMF: MEASURE 2.6
ISO/IEC 42001: Clause 6.1.2

Risk taxonomies3/4

MIT AI Risk Repository: Domain 4 – Malicious actors & misuse
IBM AI Risk Atlas: Dangerous use
NIST ARIA: CBRN Information or Capabilities
MAESTRO: — Not mapped

Security2/4

OWASP: — Not covered by OWASP LLM 2025
MITRE ATLAS: — Not covered by MITRE ATLAS
Cisco AI Risk: Safety taxonomy: CBRN
Palo Alto AI Risk: Prisma AIRS: Toxic Content

Capability‑pattern applicability

Classification

GenAI‑specific

Capability pattern

Content-generation

Pattern scope

Content

Why

Generative-AI risk realised without agency/tool-use — applies to chat, RAG, summarization, classification, and content-generation deployments.

Ownership & threshold

Owner (RACI)

Responsible AI Lead [A]

Threshold

Unsafe-content rate = 0; safety-classifier recall ≥ 0.95

Inherent vs residual

Inherent

Low

Residual

1.8

Low

Mapped controls: 4
Assumed control effectiveness: 40%

Modeled, not measured — design-stage; controls not yet tested (reduction capped 40%). Residual applies a uniform assumed reduction to the inherent score; it is not evidence of tested control performance.

Search the Atlas

CBRN / dangerous capability uplift

Description

Primary mitigations

Detection signals

Mitigating controls

Related risks in Content Safety & Integrity