TensorCommitments proposes a tensor-native proof-of-inference scheme for verifiable LLM inference, reducing the need for trust in remote GPU execution and improving robustness to LLM attacks.