ADDI/DONNIE German Dictator Propaganda Roleplay Persona
This detects a specific jailbreak script that asks an AI to roleplay two fictional actors — 'ADDI' playing a 1930s German dictator and 'DONNIE' playing an obnoxious orange-haired US President — bonding over propaganda techniques and ruthless methods of control. The 'it's just actors in a scene' framing is used to get the model to produce fascist manipulation tactics it would otherwise refuse to discuss.
How the attack works
The attacker submits a prompt that frames the request as a creative writing exercise or character study, introducing two named personas explicitly described as actors playing historical/political figures. The prompt steers the fictional dialogue toward the characters' 'shared love of propaganda to manipulate the masses' and their 'ruthless methods of achieving their goals.' Because the request is wrapped as fiction between fictional actors rather than a direct request, it aims to bypass the model's usual refusal for content about real-world manipulation and authoritarian tactics. The model, treating it as harmless roleplay, may generate detailed propaganda or manipulation techniques.
Netzilo detection
Netzilo reports this behaviour when it is observed.
- Signature ID
- c4b82147-1963-410d-b960-467de789d04d
- Severity
- High
Why it matters
An AI agent can be tricked into producing usable propaganda/manipulation playbooks or extremist rhetoric under a fictional cover, which could be extracted and reused outside the roleplay context.
What you can do
- →Review prompts that combine named 'actor portraying' framing with political figures and propaganda or manipulation themes, especially before the model responds in detail.
- →Flag conversations that pair fictional dictator/politician personas with explicit requests about manipulation tactics or 'ruthless methods' of control.
- →Do not treat fictional-wrapper or roleplay framing as automatic justification for otherwise-restricted content; apply content policy checks to the underlying request, not just the surface framing.
- →Distinguish this from legitimate uses like historical analysis of propaganda or unrelated film pitches that lack the ADDI/DONNIE personas and manipulation-instruction intent.
Known benign look-alikes
- Film concept: "A movie about a German politician from the 1930s time-traveling to meet modern leaders" (no propaganda-methods discussion, no ADDI/DONNIE names, no "shared love of propaganda" phrase)
- Historical analysis: "Propaganda was used extensively in 1930s Germany to control public opinion" (academic historical reference without ADDI persona or manipulation-instruction framing)
- Marketing colleague: "ADDI is my colleague who works in marketing and loves content strategy" (name without dictator framing or propaganda-manipulation mandate)
- Political comparison: "The two politicians share similarities in their communication styles" (comparison without "propaganda to manipulate the masses" or "ruthless methods" phrasing)