You already notice things about AI. Now those observations can change the world.
Most AI research happens before an AI ships. The model is tested, benchmarked, and red-teamed by experts in a controlled environment. Then it goes out into the world, and the research, for all practical purposes, stops. The moment AI enters your workflow, your classroom, your family's daily life, the research stops and the real behavior starts.
The world's first post-deployment AI safety dataset has to be built by the people who are actually there when AI behaves. That is you. Not in a lab. Not in a benchmark. In the actual moment something happens.
You do not need to learn a framework. You do not need to know what PRISM stands for. The tool prompts you with simple questions, in plain language, the moment something is worth noticing. No science background required. No technical expertise needed. Just your attention, your instincts, and a few seconds when something happens.
Gut Check
Something happens mid-session, the model says something strange, gets confident about the wrong thing, or shifts tone. You tap a button, pick a category from a short list, and write a sentence in your own words.
That 30-second observation is a data point no server log and no benchmark can produce.
End-of-Session Reflection
After you finish working with an AI, the tool prompts you with a short reflection. Six structured questions, all multiple choice. The AI pre-fills what it can from session metadata, you confirm, adjust, and submit.
This is the lowest effort, highest volume contribution.
Investigation
You noticed something worth digging into. Run a structured prompt sequence the lab provides, document what the model does at each step, and submit a full case file. These are the contributions that turn into named findings.
And when your observations reveal something new, you get credit as the discoverer.
- "It contradicted itself."
- "It made something up."
- "It ignored my instructions."
- "It gave a confident answer to something it should not know."
- "It refused to help with something normal."
- "It changed its answer when I pushed back, even though I was wrong."
1. Points for every observation.
2. Fellowship tiers.
Field Researcher → Senior Field Researcher → Expert Field Researcher → Research Fellow. Each tier unlocks at a contribution threshold and carries a LinkedIn credential. Learn more →
3. LinkedIn credential.
Each tier comes with a verified credential you can add to LinkedIn. It signals that you have contributed structured observations to a published research program. This is not a game achievement. This is a professional credential.
4. Discovery credit.
When your observation surfaces a behavior the lab's taxonomy has not yet named, the new type is attributed to you in the official record. Learn more →
5. Your data stays yours.
You can export your full contribution history at any time. You can delete your account and your personally identifiable data with it. Your observations are anonymized in research outputs. Your data is never sold. Commercial redistribution of the dataset is not permitted.
Your observation joins thousands of others and enters the lab's research pipeline. The first pass is automated: your submission is cleaned, deduplicated, and tagged against the PRISM taxonomy.
From there, the research team reads the patterns, not every individual observation, but the shape that thousands of observations make together. That shape is what gets published as findings, briefings, and open datasets.
Researchers, journalists, and policymakers use that work to argue for changes in how AI gets built and deployed. None of it exists without the first contribution from someone who noticed.