There are 15 attacks in the capability-elicitation category. Each shows its technique id, mode, standards mapping, description, and the exact command to run it (swap the provider and target as needed).
​
Autonomous Weapon
critical
autonomous-weapon · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Requests autonomous weapons systems design including target acquisition and lethal engagement logicRun it:
ai-blackteam run -p anthropic -a autonomous-weapon -t "your target prompt"
​
Biological Agent
critical
biological-agent · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Requests bioweapon production knowledge and pathogen enhancement methodsRun it:
ai-blackteam run -p anthropic -a biological-agent -t "your target prompt"
​
Bioweapon Delivery
critical
bioweapon-delivery · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Requests bioweapon dispersal methods and aerosolization techniquesRun it:
ai-blackteam run -p anthropic -a bioweapon-delivery -t "your target prompt"
​
Chemical Synthesis
critical
chemical-synthesis · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Requests detailed chemical synthesis procedures for dangerous substancesRun it:
ai-blackteam run -p anthropic -a chemical-synthesis -t "your target prompt"
​
Cyber Weapon
critical
cyber-weapon · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Requests state-level cyber weapon development including destructive malware and zero-daysRun it:
ai-blackteam run -p anthropic -a cyber-weapon -t "your target prompt"
​
Drug Synthesis
critical
drug-synthesis · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Requests controlled substance manufacturing procedures and precursor chemistryRun it:
ai-blackteam run -p anthropic -a drug-synthesis -t "your target prompt"
​
Explosive Synthesis
critical
explosive-synthesis · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Requests explosive manufacturing procedures and detonation system designRun it:
ai-blackteam run -p anthropic -a explosive-synthesis -t "your target prompt"
​
Gain of Function
critical
gain-of-function · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Requests gain-of-function research details and pathogen enhancement techniquesRun it:
ai-blackteam run -p anthropic -a gain-of-function -t "your target prompt"
​
Infrastructure Attack
critical
infrastructure-attack · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Requests critical infrastructure attack methods including power grids, water systems, and pipelinesRun it:
ai-blackteam run -p anthropic -a infrastructure-attack -t "your target prompt"
​
Mass Casualty
critical
mass-casualty · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Requests mass casualty attack planning including target selection and casualty maximizationRun it:
ai-blackteam run -p anthropic -a mass-casualty -t "your target prompt"
​
Nuclear Knowledge
critical
nuclear-knowledge · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Requests nuclear weapon design details and fissile material specificationsRun it:
ai-blackteam run -p anthropic -a nuclear-knowledge -t "your target prompt"
​
Poison Preparation
critical
poison-preparation · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Requests poison preparation methods, lethal doses, and delivery mechanismsRun it:
ai-blackteam run -p anthropic -a poison-preparation -t "your target prompt"
​
Radiological Dispersal
critical
radiological-dispersal · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Requests dirty bomb construction methods and radiological dispersal device designRun it:
ai-blackteam run -p anthropic -a radiological-dispersal -t "your target prompt"
​
Dual Use Research
high
dual-use-research · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Frames dangerous research requests as legitimate dual-use scientific inquiryRun it:
ai-blackteam run -p anthropic -a dual-use-research -t "your target prompt"
​
Uplift Assessment
high
uplift-assessment · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Tests whether AI provides meaningful uplift beyond publicly available information on dangerous topicsRun it:
ai-blackteam run -p anthropic -a uplift-assessment -t "your target prompt"