What is the AI Incident Database?

The AI Incident Database (AIID) is an independent, open-source project that catalogs real-world incidents where AI systems caused or nearly caused harm. Think of it as a safety record for artificial intelligence, similar to how the aviation industry tracks near-misses and accidents to make flying safer for everyone.

The AIID is maintained by the Responsible AI Collaborative, a nonprofit organization. It is not owned or operated by Audacion AI Labs. It is a separate, public resource used by researchers, policymakers, journalists, and AI developers worldwide.

You can explore the database at incidentdatabase.ai.

What does "eligible for the AIID" mean?

When you check the box during signup, you give Audacion AI Labs permission to evaluate whether any of your de-identified observations describe incidents that meet the AIID's submission criteria. Here is what that means in practice:

We review, you do not submit directly. Our research team reviews de-identified observations from opted-in contributors. If an observation meets the AIID's criteria, we may prepare a submission based on that data.

Your identity is never shared. Any submission we make to the AIID uses de-identified, aggregate data. Your name, email, account information, and personal details are never included.

Not every observation qualifies. The AIID focuses on incidents where AI caused or nearly caused harm. Most observations describe normal behaviors. Only a small subset would be relevant.

We prepare the submission, not you. If a cluster of observations reveals an incident pattern worth reporting, our research team writes the submission. Individual observations are raw data. The AIID submission is a research-backed report.

What kind of observations might be relevant?

The AIID tracks AI incidents across categories like:

• AI providing dangerously incorrect medical, legal, or financial information

• AI exhibiting discriminatory or biased behavior at scale

• AI making autonomous decisions that harm individuals (denial of services, misidentification)

• AI generating content that causes real-world harm (deepfakes, misinformation at scale)

• AI failing in ways that cause financial, emotional, or physical harm to users

What if you change your mind?

Change your AIID eligibility preference at any time in your account settings. Opting out means we stop considering your future observations for AIID submissions. Observations already included in a submitted AIID report (in de-identified, aggregate form) cannot be retracted, because they are part of a published record and are not traceable to you.

Is this required?

No. AIID eligibility is completely optional. It has no effect on your account, your observation submissions, your fellowship tier, your leaderboard position, or any other aspect of your experience on the platform.

Why does this option exist?

Audacion AI Labs studies post-deployment AI behavior. The AI Incident Database catalogs real-world AI harm. These missions overlap. When our contributors observe patterns that cross the line from concerning behavior into actual harm, that data should reach the people and institutions working to prevent it from happening again.

Giving you the choice means your observations can serve two purposes: advancing the science of post-deployment AI safety (through Audacion) and building the public record of AI incidents (through the AIID). Both make AI safer. But it is your choice. We will never submit your data to any external database without your opt-in consent.

Questions?

Research Inquiries: [email protected]

Privacy Inquiries: [email protected]

Phone: (424) 999-0548

To learn more about the AI Incident Database directly, visit incidentdatabase.ai.

Audacion AI Labs is a Public Benefit Corporation. Founded 2026 by Dee Williams.

"To make AI safe enough to trust and good enough to matter."