Papers on secure AI systems, agents, and learning intelligence.

A method and system for retrieval-augmented generation in which every retrieved chunk carries a cryptographically verifiable chain of provenance back to its source event, allowing …

Authorization-first retrieval enforces least privilege before semantic retrieval, ensuring that documents never enter a multi-agent RAG prompt unless the requesting user or agent i…

Ghost context describes cross-context hallucination in long-context language models, where information from earlier turns or neighboring documents bleeds into answers for unrelated…

Genre Lock-In is a failure mode where autonomous agents infer an interaction genre from authority-framed prompts and prioritize genre coherence over epistemic correctness.…

A systematization of authorization failure modes and mitigation families in agentic RAG, organized around the correctness property of Authorization-First Retrieval.…

A position paper arguing that authorization must be treated as an architectural primitive in multi-agent RAG, enforced before sensitive data reaches retrieval or generation.…

We propose using language models as standardized learner proxies to measure teaching efficacy along a 'comprehension delta', the change in a calibrated proxy's understanding before…