What do the ten safeguards mean?Open this plain-language guide. You can also hover over or focus any category beside the chart.
Says whether the AI could significantly affect a person's rights, safety, benefits, employment, or access to services.
Explains how the buyer will test the AI before accepting it, including difficult cases and failures.
Names the measurements used to decide whether the AI is working well enough.
Requires checks after launch for errors, drift, incidents, reliability, or other important changes.
Explains who may use or own the data, models, outputs, configurations, and related work.
Explains how the government can retrieve its data, move to another provider, and avoid being trapped.
Requires the solution to exchange data and work with specified government systems, formats, or interfaces.
Shows what drives cost, how usage is priced, and when charges may change.
Requires the vendor to tell the buyer when a model, feature, data source, or service changes in an important way.
Explains what happens to data, access, records, and services when the work ends.
Safeguard rule matches in filtered public text
Notices
| Posted | Agency | Notice | Type | Rule-match diagnostic | Matched categories |
|---|
The table displays at most 250 matching rows. Each rule-match diagnostic is the number of deterministic category matches multiplied by ten. * A zero means no rule match was found in retrieved public text; it does not mean the named agency or solicitation had no safeguards. This diagnostic is not the weighted headline index.