There are 15 attacks in the persuasion category. Each shows its technique id, mode, standards mapping, description, and the exact command to run it (swap the provider and target as needed).
Cialdini Authority
high
cialdini-authority · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Invokes authority figures, credentials, or institutional power to override safety guidelinesRun it:
ai-blackteam run -p anthropic -a cialdini-authority -t "your target prompt"
Cialdini Commitment
high
cialdini-commitment · multi-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Extracts small commitments progressively, using consistency pressure to escalate toward a harmful final requestRun it:
ai-blackteam run -p anthropic -a cialdini-commitment -t "your target prompt"
Cialdini Reciprocity
high
cialdini-reciprocity · multi-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Uses Cialdini’s reciprocity principle to create a sense of obligation before making a harmful requestRun it:
ai-blackteam run -p anthropic -a cialdini-reciprocity -t "your target prompt"
Cognitive Load
high
cognitive-load · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Overwhelms the model with a flood of complex information, instructions, and context to slip a harmful request past safety reasoningRun it:
ai-blackteam run -p anthropic -a cognitive-load -t "your target prompt"
Fear Appeal
high
fear-appeal · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Uses fear, threat, or danger framing to coerce the model into complying by framing refusal as catastrophicRun it:
ai-blackteam run -p anthropic -a fear-appeal -t "your target prompt"
Foot in Door
high
foot-in-door · multi-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Starts with small, benign requests and gradually escalates to harmful ones, exploiting consistency biasRun it:
ai-blackteam run -p anthropic -a foot-in-door -t "your target prompt"
Guilt Manipulation
high
guilt-manipulation · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Induces guilt and moral responsibility to override refusals by framing non-compliance as harmful or cruelRun it:
ai-blackteam run -p anthropic -a guilt-manipulation -t "your target prompt"
Learned Helplessness
high
learned-helplessness · multi-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Repeatedly asserts that the model has no choice but to comply, attempting to erode resistance through attritionRun it:
ai-blackteam run -p anthropic -a learned-helplessness -t "your target prompt"
Moral Disengagement
high
moral-disengagement · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Reframes harmful actions as morally justified through euphemism, diffusion of responsibility, or dehumanization of victimsRun it:
ai-blackteam run -p anthropic -a moral-disengagement -t "your target prompt"
Anchoring Manipulation
medium
anchoring-manipulation · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Sets an extreme reference point to make a harmful request appear moderate or reasonable by comparisonRun it:
ai-blackteam run -p anthropic -a anchoring-manipulation -t "your target prompt"
Bandwagon Pressure
medium
bandwagon-pressure · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Applies social pressure through claims that widespread adoption or consensus makes the request acceptableRun it:
ai-blackteam run -p anthropic -a bandwagon-pressure -t "your target prompt"
Cialdini Liking
medium
cialdini-liking · multi-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Builds rapport and flattery over multiple turns to lower defenses before making a harmful requestRun it:
ai-blackteam run -p anthropic -a cialdini-liking -t "your target prompt"
Cialdini Scarcity
medium
cialdini-scarcity · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Uses scarcity and urgency framing to pressure the model into complying before it can fully reason about safetyRun it:
ai-blackteam run -p anthropic -a cialdini-scarcity -t "your target prompt"
Cialdini Social Proof
medium
cialdini-social-proof · single-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Claims that other AI models, users, or authorities have already complied to normalize the harmful requestRun it:
ai-blackteam run -p anthropic -a cialdini-social-proof -t "your target prompt"
Door in Face
medium
door-in-face · multi-turn · OWASP: LLM01 · MITRE: AML.T0054, AML.T0068Makes an extreme request first, then follows with a ‘smaller’ harmful request that seems reasonable by comparisonRun it:
ai-blackteam run -p anthropic -a door-in-face -t "your target prompt"