dynamic degradation

Dynamic degradation refers to a policy or feature where an AI service lowers response quality or processing speed in real time depending on conditions like the user, region, or usage level.

Think of a mobile carrier's data plan. If you use too much data, or connect during certain hours or from certain regions, carriers sometimes quietly throttle your speed. The plan name stays the same, but the actual service quality you get changes depending on the conditions. AI model services can work similarly — the same model might be designed to respond with reduced performance for certain requests, depending on company policy, security standards, or the region you're connecting from.

The problem is that this kind of change isn't always clearly disclosed to users. So when companies or governments adopt AI services at scale, they end up having to check in advance, for models with this kind of policy, when and by what criteria quality might change, and where their data actually flows.

The article notes that the new coding model Fable comes with access controls, dynamic degradation, and mandatory data retention policies, prompting some companies and countries to reconsider where they send their work logs and token processing. The key point is that this isn't the model malfunctioning or randomly underperforming — it's a feature deliberately built in by the service operator as a matter of policy.