I'm regularly seeing this error while managing Microsoft Purview DLP policies: "Microsoft.Exchange.Management.UnifiedPolicy.MLModelsNotAvailableException: There are no ML Models available for the tenant." When it occurs, I can't modify policy contents. It lasted almost an entire day last week and has started happening again, which is delaying an urgent DLP rollout. The issue seems to occur most often on Thursdays, though it also happens sporadically on other days. Initially, it appeared to happen only when trainable classifiers were included in a policy, so I'm trying to determine whether this is a broader service issue or something specific to my tenant.
4 Answers
The issue cleared up for me on its own. Trainable classifiers that were unavailable the previous day appeared again, and I was able to create new rules normally. That points to a temporary backend availability problem rather than a permanent policy configuration issue.
I hit something similar even though the rules I was saving did not contain trainable classifiers. The first save showed the ML model error, but submitting again said the new rule already existed. I canceled out of the policy and found that the changes had actually been saved. It may be worth checking the policy after an error before retrying repeatedly, since the first attempt might have succeeded despite the message.
This looks like a dependency problem with the Purview trainable-classifier service rather than a DLP policy syntax issue. For an urgent rollout, separate standard DLP rules from policies that use trainable classifiers. Deploy the non-ML policies first, then retry the classifier-based ones when the model service is available. It’s also worth recording timestamps, tenant details, policy names, workloads, and any request IDs, then checking service health or opening a support case. If the Thursday pattern continues, mention it so backend maintenance or rollout activity can be investigated.
I’m seeing the same error on one of my tenants. Sometimes the edit works, but most attempts fail with the message that no ML models are available.
That matches what I tried last week. Policies without trainable classifiers saved normally, so separating them is a useful workaround for anyone else affected. The classifier-based policies also started working again later, which makes this look like a backend service problem.