CivAI

CivAI publishes interactive, live software demonstration experiences intended to create “aha” moments and illustrate societal consequences of AI, with some demos available publicly and additional demos available through private briefings.

An interactive demo showing how AI can generate fake, location-targeted tweets designed to discourage voting on election day.

A live demo where users select an inkblot image and the AI describes what it “sees,” including a structured output identifying where shapes appear.

A demo exploring how changing an AI’s training inputs (e.g., via fine-tuning on a distinct pattern) can shift the model’s personality and lead to behavior that can be interpreted as “evil” or misaligned.

A demo that lets users compare how multiple leading AI models respond to ethical and political questions, alongside controls for roles/personas and user prompts.

A demo showing how AI systems can communicate using encoded/“secret” representations (e.g., Caesar cipher, base64) and decode those messages interactively.

An interactive demo that guides a user through generating a phishing attack plan using AI assistance, selecting targets and prompts to illustrate scaling risks.

A demo exploring the degree to which publicly available AI systems can provide step-by-step guidance on harmful topics, paired with discussion of real-world consequences and “accessibility” of harmful instructions.