Who is the Agent to Blame? Localizing Faithfulness and Citation Mistakes in Agentic Deep Research

Summary

Peer-reviewed (EMNLP 2026) method for attributing faithfulness and citation errors in a multi-agent deep-research pipeline to the specific agent invocation that introduced them, with a four-type taxonomy (hallucination, uncited input reliance, uncited output, insufficient citations). Finds errors pass through the pipeline like a telephone game and that 84.7% of final-report errors in AI-Q originate at the orchestrator, ~31% of them hallucinations. Directly relevant to anyone designing a multi-worker research pipeline and needing per-stage accountability rather than end-of-pipeline evaluation.

Contribution

Queued — not yet read.

Method

Queued — not yet read.

Relevance to RISE

Surfaced by the 2026-08-29 literature sweep as a high-priority candidate for the RISE knowledge base (angle: research-agent benchmarks). Verified against: arXiv abstract page opened 2026-08-29 (title, 6 authors, submitted 25 Aug 2026, EMNLP 2026 Main note and full abstract read)

Critique / open questions

Queued — not yet assessed.

Key quotes

Deep research (DR) systems produce long-form cited reports by orchestrating multiple agents that search and synthesize information from the web. (abstract)

Citations are the primary mechanism for evaluating the faithfulness of these reports, yet current DR systems exhibit poor citation recall. (abstract)