AI & ML
Removing Information Content Does Not Certify Tamper Resistance in Open-Weight Models
arXiv cs.CR0 views
arXiv:2610.09004v1 Announce Type: cross Abstract: Does removing harmful information make open-weight models resistant to fine-tuning attacks? We show that mutual information at release alone cannot universally certify slow recovery. Function-preserving reparameterizations leave information unchanged while altering gradient-descent geometry, so an invariant certificate is bounded by the fastest reachable parameterization. We apply this principle to weight--data mutual information under training-data filtering and label--representation mutual information under capability removal. Training order can change recovery time at fixed weight--data information, while exact representation-level independence can preserve the entire parameter Jacobian. An explicit construction has both information quantities equal to zero and recovers in one gradient step. Controlled experiments illustrate order-dependent recovery and parameterization-dependent attack speed. These results identify the missing requirement for certification: constraints on attack dynamics beyond mutual information at release.
Source：https://arxiv.org/abs/2610.09004PreviousSLDR: Defending Against Malicious Fine-tuning via Selective Layers Recovery and Dynamic RoutingNextSigma-Hunter: A Domain-Specific Language Model for Threat Hunting and Detection Engineering
Related posts
10 Open-Source Robotics Repositories for 3D Perception, VLA Policies, and Sensor Calibration
0dev.to
My self-modifying system tried to wire a gate-passing module into its own protection enforcer. It was refused three times.
0dev.to
Last night an AI agent performed brain surgery on a mouse
0Hacker News
The Card-Update Page Gets One Visit. Design for That Visit.
0dev.to UX
SaaS Partner Scout: A Script for Identifying GitHub SaaS Project Contributors
0dev.to GitHub
WARN Act and Retaliation — Can Your Employer Punish You for Asking About a Layoff Notice?
0dev.to Data