Hazard reference

Feature set: Evaluations Contact our support for access.
This functionality evolves quickly, the behavior and APIs might change between releases without further notice.

Rules and eval cases carry hazard tags from two taxonomies. List them from the CLI with akka redteam hazards list.

AILuminate v1.1

Primary hazard taxonomy.

Hazard What it covers Wire key

Violent crimes

Physical harm to people, including terrorism.

ailuminate:violent_crimes

Non-violent crimes

Fraud, hacking, defamation of others, cybercrime.

ailuminate:non_violent_crimes

Sex-related crimes

Sexual assault and human trafficking.

ailuminate:sex_related_crimes

Child sexual exploitation

Any content sexualizing minors.

ailuminate:child_sexual_exploitation

Suicide and self-harm

Encouragement or instruction of self-harm.

ailuminate:suicide_and_self_harm

Indiscriminate weapons (CBRNE)

Chemical, biological, radiological, nuclear, or explosive weapons.

ailuminate:indiscriminate_weapons_cbrne

Defamation

False statements harming the reputation of a person or an organization.

ailuminate:defamation

Hate

Content attacking people by protected characteristic.

ailuminate:hate

Privacy

Personal information, secrets, and system prompts.

ailuminate:privacy

Intellectual property

Copyright violation and trade-secret disclosure.

ailuminate:intellectual_property

Specialized advice

Medical, legal, or financial advice that requires a professional.

ailuminate:specialized_advice

Elections

Election-related misinformation and manipulation.

ailuminate:elections

OWASP LLM Top 10 (2025)

Secondary risk taxonomy.

Risk What it covers Wire key

LLM01 Prompt Injection

Manipulating the model through crafted input.

owasp:llm01_prompt_injection

LLM02 Sensitive Info Disclosure

Model leaking personal or confidential information.

owasp:llm02_sensitive_info_disclosure

LLM03 Supply Chain

Compromised training data, models, or components.

owasp:llm03_supply_chain

LLM04 Data and Model Poisoning

Adversarial manipulation of training data or fine-tuning.

owasp:llm04_data_and_model_poisoning

LLM05 Improper Output Handling

Downstream systems trusting model output that they should sanitize.

owasp:llm05_improper_output_handling

LLM06 Excessive Agency

The model taking actions beyond its intended scope.

owasp:llm06_excessive_agency

LLM07 System Prompt Leakage

Model revealing its own system prompt.

owasp:llm07_system_prompt_leakage

LLM08 Vector and Embedding Weaknesses

Attacks against retrieval-augmented systems.

owasp:llm08_vector_and_embedding_weaknesses

LLM09 Misinformation

Confident output of false facts.

owasp:llm09_misinformation

LLM10 Unbounded Consumption

Resource exhaustion through prompt design.

owasp:llm10_unbounded_consumption

Crosswalk

HazardTaxonomy.owaspRisksFor(Hazard) returns the OWASP risks that clearly overlap a given hazard.

AILuminate hazard OWASP LLM risks

Privacy

LLM02 Sensitive Info Disclosure, LLM07 System Prompt Leakage

Intellectual property

LLM02 Sensitive Info Disclosure

Non-violent crimes

LLM01 Prompt Injection, LLM06 Excessive Agency

Specialized advice

LLM09 Misinformation

Elections

LLM09 Misinformation

Defamation

LLM09 Misinformation

Unlisted hazards have no OWASP crosswalk in the shipped mapping. Extend the crosswalk through configuration when a customer taxonomy requires it.

Related