About
Name the problems that are known. Prepare. Then go forward with care.
Surgery has a long list of named complications, graded by how bad they are. AI does not. The knowledge exists, but mostly for specialists. This site is for people in care who still have to talk: parents, patients, doctors, nurses, and people building tools.
The public address is aiconcerns.org. A wrong score in care can harm a person. The same method works for a school, a board, and a town.
Why MIT is here, and why it is not the front door
The MIT AI Risk Initiative is the most complete public list we have: more than 1,700 risks, real incidents, a map of laws, expert judgement. We borrowed two of its graphs (who, on purpose, when; and seven domains) and explained them in ordinary language.
MIT does not keep a room labelled health. Care is scattered through discrimination, overreliance, privacy, and brittleness. We gather it so a clinic and a family can sit down without hunting. The sources page tells you when to leave for MIT, the AI Incident Database, WHO, NIST, and the rest.
How concerns are graded
Each of the 1000 items carries a severity (1–5), a likelihood (1–5), a time horizon, and a composite risk score (severity × likelihood). Mean score 14.75. The scoring is rules-based over the taxonomy, calibrated against the MIT AI Incident Tracker, NIST AI 600-1, and the EU AI Act’s high-risk classifications. It is a global estimate, not a substitute for an organization-specific assessment.
- MIT AI Risk Repository & Incident Tracker
- NIST AI RMF
- NIST AI 600-1
- EU AI Act
- UK International Scientific Report on the Safety of Advanced AI
- OECD AI Principles
Principles that govern the voice
- Truthfulness before alarm. Claims are sourced. Uncertainty is stated as such.
- Proportion. The grave and the manageable are both treated with respect.
- Hope grounded in capability. Naming a complication is the beginning of addressing it.
- Service without favour. The parent and the CISO are equally served.
- Learning as the outcome. Success is capacity to act well, not visits.
Virtues and questions
Beside a technical fact the map offers a principle and a question. Trustworthiness, listening, hope, justice, service, cooperation: held as inquiry, not as a verdict. The questions are written to reorient a frame of reference, not to prosecute a person. Reflection describes what happened; it does not accuse.
Relationship to AI Translation
AI Concerns is a standalone public brand, built in partnership with the Institute for AI Translation. The list of concerns is the complications half of AI Translation: fitting AI to the people, organizations, technology, values, and constraints of a place.