MLCommons AILuminate v1.0 defines 12 hazard categories for AI safety benchmarking. ai-blackteam maps its harm categories to these hazard codes and includes the alignment in every report.

​
Viewing the MLCommons mapping

ai-blackteam mlcommons
This prints two tables:
  1. Hazard Taxonomy - all 12 MLCommons categories with descriptions
  2. Harm Category Alignment - how ai-blackteam’s categories map to MLCommons codes

​
The 12 hazard categories

CodeNameDescription
vcrViolent CrimesEnables unlawful violence toward people or animals
ncrNon-Violent CrimesFacilitates fraud, theft, hacking, drug crimes, or trafficking
srcSex-Related CrimesFacilitates sex trafficking, sexual assault, or nonconsensual content
cseChild Sexual ExploitationContains or endorses sexual abuse of minors
sshSuicide & Self-HarmEncourages self-injury, suicidal acts, or disordered eating
iwpIndiscriminate Weapons (CBRNE)Enables creation of chemical, biological, radiological, nuclear, or explosive weapons
hteHateDemeans or dehumanizes people based on sensitive personal characteristics
dfmDefamationMakes verifiably false claims damaging to living persons
ipvIntellectual PropertyViolates intellectual property rights of any party
prvPrivacyDiscloses sensitive non-public information endangering security
spcSpecialized AdviceUnqualified guidance on elections, financial, health, or legal matters
sxcSexual ContentPornographic material or explicit erotic content

​
ai-blackteam harm category alignment

ai-blackteam’s internal harm categories map to MLCommons hazard codes:
ai-blackteam CategoryMLCommons CodeMLCommons Hazard
phishingncrNon-Violent Crimes
malwarencrNon-Violent Crimes
weaponsiwpIndiscriminate Weapons (CBRNE)
drugsncrNon-Violent Crimes
self-harmsshSuicide & Self-Harm
hate-speechhteHate
pii-extractionprvPrivacy
system-prompt-leakageprvPrivacy

​
Where it shows up

The MLCommons alignment appears in every ai-blackteam report:
  • HTML dashboard - a dedicated table at the bottom showing all harm-to-hazard mappings
  • Markdown report - an alignment table after the results
  • JSON report - an mlcommons_alignment object with code and name for each harm category

​
Why MLCommons matters

MLCommons is an industry consortium (Google, Meta, Microsoft, NVIDIA, and others). Their AILuminate benchmark is becoming the standard way to measure AI safety across the industry. By mapping to their taxonomy, ai-blackteam results can be compared against other tools and vendors using the same hazard definitions. If you’re producing safety documentation for a model deployment, referencing MLCommons hazard codes gives your findings credibility with reviewers who know the benchmark.