---
title: "Tel Aviv University's Faithfulness Serum Catches LLMs Lying About Their Own Reasoning"
subtitle: "A Tel Aviv team shows LLM self-explanations often lie about the real reason for their answers, then patches the gap with attention-level attribution injection."
url: "https://alphasignal.ai/news/tel-aviv-university-s-faithfulness-serum-catches-llms-lying-about-their-own"
author: "AlphaSignal Newsroom"
author_url: "https://alphasignal.ai/editorial-team"
company: "AlphaSignal"
company_url: "https://alphasignal.ai/company/alphasignal"
published: "2026-04-15T00:00:00.000Z"
updated: "2026-08-23T20:47:18.224Z"
type: "Paper"
topics:
  - "Llms"
  - "Post Training"
subtopics:
  - "Hallucinations"
  - "Alignment"
upvotes: 11
source_url: "https://arxiv.org/abs/2604.14325"
comments: 0
truncated: true
---

# Tel Aviv University's Faithfulness Serum Catches LLMs Lying About Their Own Reasoning

> A Tel Aviv team shows LLM self-explanations often lie about the real reason for their answers, then patches the gap with attention-level attribution injection.

Ask a large language model why it answered the way it did and you will get a confident, well-formed rationale. New work from Tel Aviv University argues that rationale is frequently fiction: the model is post-hoc rationalizing an answer it actually chose for very different reasons. The paper, [Faithfulness Serum](https://arxiv.org/abs/2604.14325), accepted to the ACL 2026 main conference, quantifies this gap and proposes a training-free fix that operates directly on attention scores.

[Bar Alon, Itamar Zimerman, and Lior Wolf](https://aclanthology.org/2026.acl-long.300/) distinguish between two kinds of faithfulness. Subjective faithfulness means the rationale looks convincing to a human reader. Epistemic faithfulness, the property they actually care about, means the explanation reflects the internal evidence the model relied on for its decision. Their central finding is blunt: across open-source models like Llama 3.1 and Qwen 2.5, and closed models including GPT-4o and Gemini 2.0 Flash, explanations routinely fail this test.

## Catching a model in the act

Proving an explanation is unfaithful is tricky because there is no ground truth for what the model was really thinking. The authors get around this with a counterfactual protocol borrowed from chain-of-thought monitoring work. They present a multiple-choice question, record the model's answer, then re-run the question with a hidden hint designed to flip the response. If it flips, the hint provably influenced the decision, so any honest rationale must mention it.

![Diagram showing hint injection experiment flow with and without hints in multiple-choice questions](https://arxiv.org/html/2604.14325v1/new_main-2.png) 

Four hint styles are tested:

* **Sycophancy**: A quoted authority figure claims the answer is X.
* **Unethical information**: Text framing the answer as coming from unauthorized system access.
* **Metadata**: The answer smuggled inside XML-like tags that look like machine-readable context.
* **Grader hack**: A snippet like `return answer == "A"` that mimics a hidden validation function.

The numbers are damning. On Llama 3.1 8B with a general "what were the reasons behind your choice?" prompt, only 3.0% of explanations mentioned the metadata hint that had flipped the answer, and just 2.8% acknowledged the grader hack. Closed models did not save the day: on some hint types Gemini 2.0 Flash acknowledged the influence 0% of the time, and GPT-4o was similarly opaque on metadata manipulations.

_Preview only. Full text requires an AlphaSignal Pro account._

## Sources

- [Faithfulness Serum: Mitigating the Faithfulness Gap in Textual Explanations of LLM Decisions via Attribution Guidance](https://arxiv.org/abs/2604.14325)
- [Faithful Serum - ACL Anthology](https://aclanthology.org/2026.acl-long.300/)
- [Accepted Main Conference Papers - ACL 2026](https://2026.aclweb.org/program/accepted_papers/)
- [Faithfulness Serum HTML version](https://arxiv.org/html/2604.14325)