tensorflow is vulnerable to denial of service. An attacker can force a node to have, as input, a tensor backed by a nullptr
buffer via a malicious TFLite model, eventually converting the tensors to be one of read-write via a null pointer dereference.