Frequently asked questions

Audacion AI Labs is an independent post-deployment AI safety research lab. We study how AI systems actually behave once they are in the world and being used by real people. Most safety research happens before a model is released. We focus on what happens after.

It means we watch AI systems after they have been released to the public. Every conversation, every weird answer, every helpful moment is data the labs that built the model usually do not see. Citizens who use AI every day see it first. We help them log what they see so researchers can study it.

No. We are an independent Delaware Public Benefit Corporation. We study every major model and we do not take money from any AI lab in a way that compromises the research.

An observation is a short record of something you noticed an AI system do. It could be a strange answer, an unexpected refusal, a moment of unusual insight, a hallucination, a bias, or a behavior you cannot quite name. You write it down, tell us which AI it was and roughly when it happened, and submit. That is it.

No. The whole point is that the people noticing AI behavior in the wild are everyday users. If you can describe what you saw in plain words, you can contribute.

One date is when the AI behavior actually happened to you, even if that was last week or last month. The other is the moment you recorded it with us. The first date is the research data. The second is the audit trail. We never collapse them into one because honest research needs both.

EOT stands for End of Thread. When a conversation with an AI ends, an EOT reflection is a short note about how the whole exchange felt overall. Did it stay useful? Did it drift? Did it agree with you too much? These reflections add a different kind of signal than single-moment observations.

It is our growing library of named AI behaviors, organized into five pillars (Persuasion, Reliability, Identity, Safety, and Memory; we call this PRISM). Each behavior has a stable code like I-06, a plain-language definition, and examples. When you log an observation, you tag it with the behavior it matches.

That is called a discovery, and it is one of the most valuable things you can submit. You flag the observation as new and walk through a short guided wizard: we show you the closest existing behaviors first (so you can decide whether yours really is new), then ask you to build the case with evidence (screenshots, video, or a paste of the conversation), explain why it does not fit any existing behavior, pick the closest pillar, and confirm when it happened. Our research team reviews it. If they confirm it is genuinely new, it becomes an official behavior in the catalog with your name on it.

Yes. Discovery attribution is permanent. Your contribution is recorded against the behavior with the date you originally observed it. You can choose whether your name appears publicly.

A pattern is something the research team is actively watching. It usually spans more than one behavior and shows up across more than one AI model. Patterns have urgency levels so you can see what the team thinks needs the most attention right now.

Yes. Open your profile and use Permanently delete account. Your personal information, your observations, your EOT reflections, and your discovery submissions are removed. Two exceptions, for good reasons:

• Harm reports you filed stay on file for safety and legal reasons, but the reporter is anonymized so they are no longer linked to you.
• If you have authored a published blog post on the site, the deletion is blocked until the post is reassigned or unpublished. This only affects researchers and admins. You will see a clear message in the app telling you what to do.

The action cannot be undone.

Yes. Your profile has an export option that gives you a complete file of your observations, EOT reflections, and account information.

Every observation is stripped of personally identifying information before it enters the research corpus. Your name, email, and account identifiers are kept on the account record and are never published alongside your observation. See our Privacy Policy for the full picture.

You earn points for submitting observations, EOT reflections, referring other citizens, and having discoveries accepted into the catalog. As your point total grows, your fellowship tier advances (Field Researcher, Senior Field Researcher, and beyond). Tiers are recognition, not gatekeeping; every contribution matters.

Use the Report AI Harm form. It is separate from regular observations and can be filed anonymously. The description and any contact email you provide are encrypted at rest. Every harm-report response also includes curated crisis resources, regardless of whether you asked for them.

Please reach a human first. In the United States, call or text 988 for the Suicide and Crisis Lifeline. Outside the US, your local emergency number is the right starting point. We are a research lab, not a crisis service.

Not in a way that identifies you. Aggregated and de-identified safety findings may be published as part of research output, but never your name, email, or anything that could be tied back to you.

The best way is to contribute observations regularly. If you also want to support financially, see the Donate page.

Use the Contact page or email [email protected].

Send it to [email protected] and we will answer you directly. If we get the same question often, we will add it here.