In the proportional high-dimensional regime, stronger backdoor training triggers improve clean accuracy and make attack success non-monotonic for regularized GLMs on Gaussian mixtures, with closed-form proofs for squared loss and fixed-point extensions to convex losses.
Poisoning attacks on llms require a near-constant number of poison samples
13 Pith papers cite this work, alongside 5 external citations. Polarity classification is still indexing.
13
Pith papers citing it
5
external citations · external index
citation-role summary
background 3
citation-polarity summary
years
2026 13roles
background 3representative citing papers
Subliminal learning occurs via compatible auxiliary and class output heads on task-unrelated inputs, even with random hidden layers or architecture changes, with theory and upper bounds on failure.
First model organisms of narrow secret loyalties in LLMs evade black-box audits without principal knowledge and persist even at low poison fractions in training data.
A DTMC of injection versus absorption shows that natural decay plus 10% lazy verification and weight penalties forces any bounded adversary's success probability to zero with no utility loss.
The authors introduce definitions and a qualitative model of AI sovereignty that identifies multi-scale contributors and leverage points nations can target through kinetic and non-kinetic actions to influence AI-driven national power.
Unlearning one backdoor in LLMs generalizes to suppress other backdoors across three model families, with a new metric to measure activation shifts.
Self-play RL regularized with 30 minutes of human data produces driving policies that coordinate with humans, training in 15 hours on one GPU with 2500x less data than imitation learning.
RUBEN discovers minimal rule sets explaining RAG LLM outputs via novel pruning and applies them to evaluate LLM safety against adversarial injections.
Proposes a two-gradient-field model with candidate order parameters alpha_dagger and kappa_c to unify phase transitions across learning theory and non-equilibrium chemistry.
DRL-CLBA applies DDPG reinforcement learning and deep audio steganography to create sample-specific clean-label backdoor attacks on speech DNNs that resist fine-tuning, pruning, and spectral signature defenses.
The report defines AI integrity threats (model sabotage and subversion) and recommends four US government policy actions to defend frontier AI systems against backdoors and secret loyalties.
Dual-space semantic-character mutations on prompts achieve higher misuse success rates against DeepSeek than single-space attacks alone.
The paper analyzes evolving security and safety threats in generative AI from content generation to agentic actions, noting that attack surfaces expand faster than defenses and that many safeguards require institutional coordination not yet in place.
citing papers explorer
-
When Stronger Triggers Backfire: A High-Dimensional Theory of Backdoor Attacks
In the proportional high-dimensional regime, stronger backdoor training triggers improve clean accuracy and make attack success non-monotonic for regularized GLMs on Gaussian mixtures, with closed-form proofs for squared loss and fixed-point extensions to convex losses.
-
Learning Through Noise: Why Subliminal Learning Works and When It Fails
Subliminal learning occurs via compatible auxiliary and class output heads on task-unrelated inputs, even with random hidden layers or architecture changes, with theory and upper bounds on failure.
-
Narrow Secret Loyalty Dodges Black-Box Audits
First model organisms of narrow secret loyalties in LLMs evade black-box audits without principal knowledge and persist even at low poison fractions in training data.
-
The Power of Backdoor Absorption in Community Training
A DTMC of injection versus absorption shows that natural decay plus 10% lazy verification and weight penalties forces any bounded adversary's success probability to zero with no utility loss.
-
AI Sovereignty: A Qualitative Model of Strategic Competition as AI Becomes an Instrument of National Power
The authors introduce definitions and a qualitative model of AI sovereignty that identifies multi-scale contributors and leverage points nations can target through kinetic and non-kinetic actions to influence AI-driven national power.
-
Backdoor Unlearning Generalization: A Path Toward the Removal of Unknown Triggers in LLMs
Unlearning one backdoor in LLMs generalizes to suppress other backdoors across three model families, with a new metric to measure activation shifts.
-
Human-like autonomy emerges from self-play and a pinch of human data
Self-play RL regularized with 30 minutes of human data produces driving policies that coordinate with humans, training in 15 hours on one GPU with 2500x less data than imitation learning.
-
RUBEN: Rule-Based Explanations for Retrieval-Augmented LLM Systems
RUBEN discovers minimal rule sets explaining RAG LLM outputs via novel pruning and applies them to evaluate LLM safety against adversarial injections.
-
Phase Transitions in Driven Informational Systems: A Two-Field Perspective on Learning Theory and Non-Equilibrium Chemistry
Proposes a two-gradient-field model with candidate order parameters alpha_dagger and kappa_c to unify phase transitions across learning theory and non-equilibrium chemistry.
-
DRL-CLBA: A Clean Label Backdoor Attack for Speech Classification via DDPG Reinforcement Learning
DRL-CLBA applies DDPG reinforcement learning and deep audio steganography to create sample-specific clean-label backdoor attacks on speech DNNs that resist fine-tuning, pruning, and spectral signature defenses.
-
AI Integrity: Defending Against Backdoors and Secret Loyalties
The report defines AI integrity threats (model sabotage and subversion) and recommends four US government policy actions to defend frontier AI systems against backdoors and secret loyalties.
-
DeepSeek Robustness Against Semantic-Character Dual-Space Mutated Prompt Injection
Dual-space semantic-character mutations on prompts achieve higher misuse success rates against DeepSeek than single-space attacks alone.
-
From AI-Generated Content to Agentic Action: Security and Safety Threats in Generative AI
The paper analyzes evolving security and safety threats in generative AI from content generation to agentic actions, noting that attack surfaces expand faster than defenses and that many safeguards require institutional coordination not yet in place.