Origin Is All You Need: Provenance-Aware Transformers for Structural Trust-Boundary Separation
The paper proposes Provenance-Aware Transformers, an LLM architecture for defending against indirect prompt injection by giving every token an explicit source “ring” and using origin embeddings, attention biases, and scaling so the model can structurally distinguish system/developer authority from user or retrieved-document data. The interesting bit is that it tries to move trust separation into the transformer itself rather than relying on prompt wording or post-hoc filters, though its effectiveness likely hinges on correct provenance labels from the surrounding application, extra fine-tuning, and evaluations that would need to prove the boundary holds under realistic tool/RAG pipelines.
Abstract
Indirect prompt injection (IPI) remains a central safety and security challenge for large language model (LLM) systems because standard transformers lack architectural notion of source authority. Retrieved documents, user inputs, and system instructions are all processed through the same undifferentiated attention mechanism, forcing the model to infer from wording alone what should be obeyed and what should be treated as data. We propose Provenance-Aware Transformers, a provenance-aware defense that makes application-supplied source labels actionable inside the model. Each input token is assigned a ring ID encoding its origin, and the model is augmented with origin embeddings, a learnable origin attention bias, and a learnable origin scale that preserves provenance under normalization. The resulting architecture enforces a structural boundary between authoritative and non-authoritative sources during generation. To instantiate this architecture on released pretrained models, we propose a two-stage fine-tuning pipeline to teach the model origin semantics and task behavior under ring constraints. Evaluation shows that Provenance-Aware Transformers maintain robust resistance to IPI both in-distribution and out-of-distribution while preserving utility comparable to the base pretrained model. More broadly, our work shows that exposing provenance as a first-class architectural signal can shift LLM safety alignment from brittle pattern matching toward explicit trust separation.