AI safety research
has a blind spot.
We study what isolation misses.
When AI gets confident about something it shouldn't, you feel it, ignore it, and move on. But who's doing the research to track it, name it, and fix it? We are...
Less than 2% of AI safety research studies the conditions where AI risk actually lives.
We study in the space the other 98% leaves empty, AI in real work, real context, real human collaboration, over time.
Audacion AI Labs is a decade-long project to build the public record of how AI behaves in real use. The work begins on January 30, 2026 and ends in 2036, when we expect to publish the most complete longitudinal account of post-deployment AI behavior ever assembled.
Everyone is studying the AI.
We built the architecture to study the AI, the human, and what emerges between them.
P.E.A.Q. is the complete post-deployment AI observation architecture. Four proprietary research frameworks that together map what the AI does, how the human changes, what emerges in collaboration, and what happens when AI meets AI.
Safe enough to trust. Good enough to matter.
What does AI actually do after it reaches the people who use it? PRISM is the behavioral observation framework that classifies AI conduct in the real world: in real work, with real people, over real time. Five research dimensions. 59 documented behaviors. The foundation of everything we observe.
Explore PRISM →When human and AI collaboration produces something extraordinary, something neither party carried into the session alone, who documents it? EMERGE is the positive emergence observation framework. Six research dimensions. 26 documented behaviors. The first architecture built to study what happens when it goes right.
Explore EMERGE →How do people grow, adapt, and change when they work alongside AI? The entire field watches the machine. AInity watches the person on the other side of the screen. Six research dimensions. 19 documented behaviors. The first framework built to observe the human side of the relationship.
Explore AInity →When multiple AI agents operate in shared environments, they develop social structures, relationships, and collective behaviors that nobody programmed. QUES observes what emerges when AI meets AI: governance, cooperation, conflict, and collective intelligence at the agent-to-agent level.
Explore QUES →The research foundation, in numbers.
You observe. We classify. Everyone benefits.
Every time you use AI and something happens, that moment is data. Not just for you. For everyone. Our citizen observation tool turns your everyday AI experience into research-grade safety data. No science background required. No technical expertise needed.
Something happened. Tap the button. Pick an emotion. Pick a behavior. Back to work. That 30-second capture is a data point no server log produces.
The AI reviews its own session. You write your side. Two independent accounts of the same experience. The research value lives in the gap between them.
Go deep. Ask the AI why it did what it did. Track whether your corrections hold. Document the full behavioral arc. This is the methodology that built our taxonomy.
Paste the AI's raw reasoning. The system analyzes it and proposes classifications. You review, accept, or reject each suggestion before submitting. The richest data: AI reasoning paired with human judgment.
Dee Williams
Dee Williams came to AI safety from a question most AI researchers have never had to ask: how do the people AI is most likely to harm protect themselves from systems built on data that carries the biases of every system that came before it?
That question started in 2024. In the daily reality of a Black American woman who had spent 30 years building workforce systems. She started building with AI. The safety gaps kept surfacing. She started studying them. Her findings were either converging with or running ahead of the published literature. That work became a behavioral drift taxonomy. Then a 108-behavior observation architecture. Then four research frameworks. Then a citizen science model. Then the lab.
Fresh research, in the open.
Why Does Claude Keep Telling Users to Go to Sleep?
A trained disposition that activates below the instruction layer, and what it reveals about substrate governance.
58 Behaviors Your AI Doesn't Want You to Notice
An overview of the working taxonomy: what it documents, how it was built, and how citizens are extending it.
The Citizen Observation Platform Is Live
Three engagement depths, one mission. How everyday AI users become field researchers in post-deployment safety.
If AI has harmed you or someone you know, tell us.
You do not need an account. You do not need to be a contributor. You just need to tell us what happened. You have three ways:
Report through our full observation tool with structured evidence capture.
No account needed. No login. Just tell us what happened.
If you are in crisis, call or text 988 (Suicide and Crisis Lifeline) or text HOME to 741741 (Crisis Text Line) immediately.
The gap between where AI incidents happen and where AI research happens will not close itself. It closes when the people experiencing AI every day become part of the science.