analiza aiqbench
This research investigates how information leaks from vision-language models through their output logits, demonstrating that even the most accessible prediction layers can expose task-irrelevant data comparable to deeper internal representations. The findings highlight a critical security vulnerability where standard model outputs may inadvertently reveal sensitive inputs assumed to be hidden by the system's architecture. For practitioners deploying large models, this underscores the necessity of rigorous auditing beyond just final generation quality to assess potential side-channel information leakage.
What do your logits know?
arXiv:2604.09885v2 Announce Type: replace Abstract: Recent work has shown that probing model internals can reveal a wealth of information not apparent from the model generations. This poses a risk of unintentional or malicious information leakage, where model users are able to learn information that the model owner assumed was inaccessible. Using vision-language models as a testbed, we present the first systematic comparison of information retained at different representational levels as it is compressed from the rich information encoded in the residual stream through two natural bottlenecks: low-dimensional projections of the residual stream obtained using tuned lens, and the final top-k logits most likely to impact model's answer. We show that even easily accessible bottlenecks defined by the model's top logit values can leak task-irrelevant information present in an image-based query, in some cases revealing as much information as direct projections of the full residual stream.
Title:What do your logits know?
Abstract:Recent work has shown that probing model internals can reveal a wealth of information not apparent from the model generations. This poses a risk of unintentional or malicious information leakage, where model users are able to learn information that the model owner assumed was inaccessible. Using vision-language models as a testbed, we present the first systematic comparison of information retained at different representational levels as it is compressed from the rich information encoded in the residual stream through two natural bottlenecks: low-dimensional projections of the residual stream obtained using tuned lens, and the final top-k logits most likely to impact model's answer. We show that even easily accessible bottlenecks defined by the model's top logit values can leak task-irrelevant information present in an image-based query, in some cases revealing as much information as direct projections of the full residual stream.
Artificial Intelligence (cs.AI)
arXiv:2604.09885 [cs.AI]
(or arXiv:2604.09885v2 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2604.09885 Focus to learn more arXiv-issued DOI via DataCite
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.
Text integral sub licența deschisă a publicației · 237 cuvinte
Subiecte conexe
Challenge: Hand coding weights for efficient sequence memorisation
AI Alignment Forum · 23 iul. 2026
Evaluating Red Team and Blue Team Capability for AI Control Research
LessWrong · 24 iul. 2026
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
LessWrong · 8 oct. 2026
What We Learned Trying to Catch AI Liars: An Aletheia's Quest Retrospective
EleutherAI · 25 aug. 2026