CVE-2020-15206: TensorFlow: SavedModel protobuf DoS in inference serving

In Tensorflow before versions 1.15.4, 2.0.3, 2.1.2, 2.2.1 and 2.3.1, changing the TensorFlow's `SavedModel` protocol buffer and altering the name of required keys results in segfaults and data corruption while loading the model. This can cause a denial of service in products using `tensorflow-serving` or other inference-as-a-service installments. Fixed were added in commits f760f88b4267d981e13f4b302c437ae800445968 and fcfef195637c6e365577829c4d67681695956e7d (both going into TensorFlow 2.2.0 and 2.3.0 but not yet backported to earlier versions). However, this was not enough, as #41097 reports a different failure mode. The issue is patched in commit adf095206f25471e864a8e63a0f1caef53a0e3a6, and is released in TensorFlow versions 1.15.4, 2.0.3, 2.1.2, 2.2.1, or 2.3.1.

An attacker with access to a model registry (compromised CI/CD credential, insider, or misconfigured public bucket) crafts a TensorFlow SavedModel with deliberately malformed protobuf keys — removing or renaming required fields. The malicious model is pushed to the production model registry and picked up by tensorflow-serving during a scheduled model refresh or hot-reload. On load, TF dereferences a null or corrupt pointer from the malformed protobuf, triggering a segfault that crashes the serving process. In a kubernetes environment without proper restart policies, this takes the inference endpoint offline; with restart policies, the attacker can trigger repeated crashes to sustain denial of service. No network authentication, no ML expertise — just knowledge of the SavedModel protobuf schema.