III
AI Watch Lab
Rigorous liability testing and certification for enterprise and agentic AI — and the workforce question those systems raise: how the people whose jobs they touch are protected, retrained, and kept in the room.
The AI Watch Lab is a research program in systemic safety — the failure modes that emerge not from a single model but from the interaction of agentic systems, the data they consume, and the institutions that deploy them.
Its outputs are deliberately practical: liability frameworks suitable for adoption by insurers and regulators, certification protocols for enterprise and agentic AI, and red-team methodologies that survive peer review. The Lab also maintains a public registry of incidents and near-misses — a clearinghouse for the failure data the field, in its current state, otherwise discards.
But safety is not only a property of the system; it is a question about the people standing next to it. The Lab studies how AI enters the workforce — where it augments human judgment and where it quietly replaces it — and what a credible, humane transition owes the workers whose roles it changes or removes.
That second line of work is the Lab’s own: open methods for measuring displacement before it arrives, retraining and adoption frameworks built with workers rather than imposed on them, and an evidentiary record clear enough for a regulator, an employer, or a union to act on.
What the program is asking.
01
What evidence does an enterprise need before delegating a high-stakes decision to an autonomous agent — and who is liable when the evidence was wrong?
02
How do we certify behavior that depends on context the certifier cannot fully observe?
03
When automation removes a role, what does a credible, humane transition owe the worker who held it — and can we measure whether it was delivered?
Proposals welcome.
The Foundation accepts grant proposals, partnership inquiries, and fellowship applications from researchers whose work intersects with this program.