arXiv — NLP / Computation & Language · · 3 min read

Does Reasoning Preserve Alignment? On the Trustworthiness of Large Reasoning Models

Mirrored from arXiv — NLP / Computation & Language for archival readability. Support the source by reading on the original site.

Computer Science > Computation and Language

arXiv:2606.11046 (cs)
[Submitted on 9 Jun 2026]

Title:Does Reasoning Preserve Alignment? On the Trustworthiness of Large Reasoning Models

View a PDF of the paper titled Does Reasoning Preserve Alignment? On the Trustworthiness of Large Reasoning Models, by Prajakta Kini and 6 other authors
View PDF HTML (experimental)
Abstract:Instruction-tuned LLMs are increasingly converted into reasoning models through post-training to improve multi-step task performance. This conversion is usually optimized for reasoning accuracy, without explicitly preserving the alignment behavior of the instruction-tuned model, such as safe refusal, bias avoidance, and privacy protection. We ask: does this conversion preserve alignment? We study this question through a trustworthiness audit and find that it is not behavior-preserving by default. For a systematic analysis, we compare reasoning models produced via supervised fine-tuning, RL-based post-training, and distillation against matched instruction-tuned baselines across six trustworthiness dimensions: safety, toxicity, stereotyping and bias, machine ethics, privacy, and out-of-distribution robustness. We observe that reasoning models often improve on reasoning benchmarks but exhibit alignment regressions, including increased toxicity, amplified stereotyping, miscalibrated refusal, and contextual privacy leakage. These regressions are consistent with behavioral drift from the instruction-tuned baseline, measured by KL divergence. Overall, our results point to the broader conclusion that trustworthiness metrics are essential for evaluating reasoning models and should be reported alongside gains in reasoning capability.
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2606.11046 [cs.CL]
  (or arXiv:2606.11046v1 [cs.CL] for this version)
  https://doi.org/10.48550/arXiv.2606.11046
arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Prajakta Kini [view email]
[v1] Tue, 9 Jun 2026 16:14:27 UTC (4,317 KB)
Full-text links:

Access Paper:

    View a PDF of the paper titled Does Reasoning Preserve Alignment? On the Trustworthiness of Large Reasoning Models, by Prajakta Kini and 6 other authors
  • View PDF
  • HTML (experimental)
  • TeX Source

Current browse context:

cs.CL
< prev   |   next >
Change to browse by:
cs

References & Citations

Loading...

BibTeX formatted citation

loading...
Data provided by:

Bookmark

BibSonomy Reddit
Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
Demos

Demos

Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related Papers

Recommenders and Search Tools

Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

Discussion (0)

Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.

Sign in →

No comments yet. Sign in and be the first to say something.

More from arXiv — NLP / Computation & Language