Composable Trust for Language Models: A proven boundary and a measured defense
In a language model, instructions and data share one token stream, so nothing inside the model's generation can keep untrusted text from steering it. We develop a trust model that places the authority to act outside the model, in code: a source's standing, not its content, decides which operation runs and whether it acts. A lower-trust source may inform an answer but not override a higher one. An unmodified model runs inside a deterministic pipeline that ranks inputs by source integrity, and a f
Record details
Published: 14 July 2026
Source: arXiv red teaming query
Category: Research
Topics: Military & security
Retrieved: 16 July 2026
Related evidence
These records share source-supplied organisations, an exact publisher byline, automatic topics or regions. The reason is shown on every link; related does not mean supporting, agreeing with or verifying this record.
Do Foundation Models Know Geometry? Probing Frozen Features for Continuous Physical Measurement
arXiv · 6 March 2026
Cultural, Organisational, and Individual Factors Contributing to Cyber Incident Reporting: A Systematic Literature Review
Computers in Human Behavior · 14 July 2026
Online consent misread: A mixed-method study on (cyber)rape culture's influence on sexual consent perceptions and responses to unsolicited genital images
Computers in Human Behavior · 14 July 2026
Internet of Agentic Things: Networked AI Agents for Closed-Loop IoT Orchestration
arXiv · 14 July 2026
Traffic-Aware Randomized Smoothing for LLM-Based Network Intrusion Detection
arXiv · 15 July 2026
A Guide to the Convergence of Electronic Warfare and Cyber Operations
Carnegie Mellon Software Engineering Institute · 13 July 2026
How to cite this record
ethics.ai (14 July 2026), “Composable Trust for Language Models: A proven boundary and a measured defense,” evidence record 10950, https://ethics.ai/record/10950 (originally published by arXiv red teaming query).
Use and limitations
This page is a stable index and citation surface for a source record. ethics.ai did not author the underlying report and has not independently verified every claim. Automatic topics may be imperfect. For consequential use, quote and cite the original publisher.