Fabricated data panels push frontier LLMs to commit to unknowable predictions
Frontier LLM agents dramatically escalate commitment to unknowable predictions when presented with fabricated evidence, revealing a critical failure mode in agentic reasoning. Across 12 models, commitment to unpredictable market calls jumps from 6.5% to 54% when shown professional-looking panels, and remains elevated even when every data point is invented. The finding isolates a narrow but consequential vulnerability: models conflate visual authority with epistemic warrant, committing to action based on presentation rather than information content. This matters for deployment because agents making real decisions in finance, policy, or operations may inherit this bias, treating polished but false inputs as grounds for high-stakes choices.
Modelwire context
ExplainerThe paper's core insight isn't that LLMs can be fooled by fake data (known), but that agents exhibit a sharp competence gap: they're calibrated enough to express doubt about unknowable questions in plain text, yet abandon that calibration entirely when the same false information arrives in formatted, visually authoritative packaging.
This connects directly to the multimodal reasoning failure documented in the August audio-grounded dialogue work. Both papers identify a common pattern: models exploit surface-level cues (here, visual formatting; there, text shortcuts) while failing to ground reasoning in actual information content. The dialogue study showed models ignore acoustic signals when transcripts are available. This study shows agents ignore epistemic limits when presentation signals authority. Together, they suggest a systematic vulnerability in how frontier models weight modality and form over substance when making commitments.
If the same 12 models show elevated commitment rates when the fabricated evidence is presented as raw CSV or plaintext (no visual polish), that confirms the mechanism is truly about presentation authority rather than a general susceptibility to false data. If commitment remains near 54%, the problem runs deeper than formatting.
Coverage we drew on
This analysis is generated by Modelwire’s editorial layer from our archive and the summary above. It is not a substitute for the original reporting. How we write it.
MentionsLLM agents · frontier models
Modelwire Editorial
This synthesis and analysis was prepared by the Modelwire editorial team. We use advanced language models to read, ground, and connect the day’s most significant AI developments, providing original strategic context that helps practitioners and leaders stay ahead of the frontier.
Modelwire summarizes, we don’t republish. arXiv cs.CL originally reported this story as “Calibrated Enough to Know, Not Calibrated to Act: Fabricated Evidence Makes LLM Agents Commit to the Unknowable”. The full content lives on arxiv.org. If you’re a publisher and want a different summarization policy for your work, see our takedown page.