LmCast :: Stay tuned in

The Implications of Linguistic Illegibility for LLM Security

Recorded: Sept. 18, 2026, 8 p.m.

Original Summarized

[2609.02852] The Implications of Linguistic Illegibility for LLM Security

Skip to main content

Search

Submit
Donate

Log in

Search arXiv

Press Enter to search · Advanced search

Computer Science > Machine Learning

arXiv:2609.02852 (cs)

[Submitted on 2 Sep 2026]
Title:The Implications of Linguistic Illegibility for LLM Security
Authors:James Mickens View a PDF of the paper titled The Implications of Linguistic Illegibility for LLM Security, by James Mickens
View PDF
HTML (experimental)

Abstract:LLMs are trained to generate natural language. However, various strands of evidence indicate that an LLM's externalized linguistic outputs and mechanistically-extracted linguistic features can be an unreliable lens for understanding internal model computation. We introduce the term ``linguistic illegibility'' to broadly refer to scenarios in which an LLM's externalized or mechanistically-probed language artifacts fail to represent how the model actually thinks. We argue that the specter of linguistic illegibility is unavoidable for LLMs whose internal computations are not directly expressed via language, but rather math over activation spaces (with lossy translations between activation spaces and natural language happening at the bookends). If linguistic illegibility is always possible, then security mechanisms that rely on a model's linguistic self-reporting (e.g., chain-of-thought monitoring, constitutional self-critique, activation probing for linguistically-defined feature vectors) can never be completely sound; the model sandbox will always need isolation techniques whose guarantees do not depend on reading a model's linguistic state at all. We argue that observing a model's outputs using taint tracking is a promising approach for an effective sandbox: regardless of how a model linguistically self-reports, a taint tracking policy can define, a priori, various pieces of system state that should never be influenced by model-produced data. We also discuss several additional sandboxing mechanisms (e.g., robust virtualization, third-party auditing of sandboxing configurations) which collectively provide a critical floor beneath linguistic monitoring, and would have mitigated recent sandbox exploits by frontier models.

Subjects:

Machine Learning (cs.LG); Cryptography and Security (cs.CR)

Cite as:
arXiv:2609.02852 [cs.LG]

 
(or
arXiv:2609.02852v1 [cs.LG] for this version)

 
https://doi.org/10.48550/arXiv.2609.02852

Focus to learn more

arXiv-issued DOI via DataCite (pending registration)

Submission history From: James Mickens [view email] [v1]
Wed, 2 Sep 2026 17:37:22 UTC (33 KB)

Full-text links:
Access Paper:

View a PDF of the paper titled The Implications of Linguistic Illegibility for LLM Security, by James MickensView PDFHTML (experimental)TeX Source

view license


Current browse context:
cs.LG

< prev

  |  
next >

new
|
recent
| 2026-09

Change to browse by:

cs
cs.CR

References & Citations

NASA ADSGoogle Scholar
Semantic Scholar

export BibTeX citation
Loading...

BibTeX formatted citation
×

loading...

Data provided by:

Bookmark

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?)

IArxiv recommender toggle

IArxiv Recommender
(What is IArxiv?)

Author
Venue
Institution
Topic

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.

Which authors of this paper are endorsers? |
Disable MathJax (What is MathJax?)

We gratefully acknowledge support from
our major funders,
member institutions, ,
and all contributors.

About
·
Help
·
Contact
·
Subscribe
·
Copyright
·
Privacy
·
Accessibility
·
Operational Status (opens in new tab)

Major funding support from

The work introduces the concept of linguistic illegibility to describe situations where an LLM's externalized or mechanistically-probed language artifacts do not accurately reflect the model's internal computational processes. The authors argue that this linguistic illegibility is inherent for LLMs whose internal computations are fundamentally represented by mathematics over activation spaces, where the translation between these activation spaces and natural language involves lossy mappings at the boundaries.

This fundamental disconnect between internal mathematical computation and external linguistic expression has critical implications for the security of large language models. If linguistic illegibility is a persistent possibility, security mechanisms that depend on a model's linguistic self-reporting, such as chain-of-thought monitoring, constitutional self-critique, or activation probing based on linguistically defined feature vectors, cannot be considered entirely sound. Consequently, a robust model sandbox must implement isolation techniques whose guarantees are independent of reading or interpreting the model's linguistic state.

The authors propose an alternative approach for effective sandboxing by focusing on taint tracking of model outputs. They contend that regardless of how a model linguistically reports its state, a taint tracking policy can define specific segments of system state that must remain inviolable, preventing them from being influenced by model-produced data. Furthermore, the paper discusses several complementary sandboxing mechanisms, including robust virtualization and third-party auditing of sandboxing configurations, which collectively establish a necessary baseline layer beneath linguistic monitoring. The authors suggest that these combined approaches could have effectively mitigated recent sandbox exploits encountered by frontier models.