The paper demonstrates that using raw source text for fine-tuning LLMs on vulnerability detection causes high false-positive rates by memorizing surface-level syntax, a problem mitigated by using Abst…

This paper investigates how reliable large language models are in recognizing their own compromised outputs during adversarial prefill attacks in safety contexts.

This paper proposes a method to recover recoverability structure from failed traces of post-trained language models, enabling test-time routing and post-training analysis.

The paper investigates the fragility and recovery mechanisms of chain-of-thought (CoT) answer hijacking, demonstrating that specific problem cells are susceptible to targeted recovery and that source…

The paper introduces Responsible Contrastive Soft Prompting (RCSP), a parameter-efficient method using soft prompts to improve LLM reliability by simultaneously suppressing hallucinations, encouraging…

This paper systematically diagnoses the failure modes of linear deception probes in LLMs, finding that while single-direction probes are insufficient, multi-dimensional probes can recover robust detec…

This paper shows that Frontier LLMs can perform multi-step reasoning over filler tokens, and introduces an unsupervised decoding pipeline to recover intermediate values.

This paper explores the vulnerability of post-hoc explanation methods in detecting audio deepfakes using a psychoacoustic framework that optimizes inaudible perturbations.

The paper introduces a novel framework to quantify faithful confidence expression (FC) in Large Reasoning Models (LRMs), finding that FC remains a significant and challenging reliability target for th…

The study demonstrates that robust, domain-invariant representations of synthetic deception can be rapidly entrenched in LLMs using modest fine-tuning, detectable by linear probes even in early layers…

The paper introduces Probe-Geometry Alignment (PGA), a surgical method that removes the measurable cross-sequence memorization signature from large language models without degrading their general capa…

The paper proposes using an auxiliary reconstruction task, specifically one that captures intra-state feature dependencies, to improve the quality of state representations learned by the encoder in ne…

Weak self-training on synthetic data can amplify a language model's existing capabilities, but this effect is strictly dependent on the compatibility between the source and student models, not on the…

The paper demonstrates that refusal behavior in Large Language Models (LLMs) is encoded as an actionable, linearly decodable signal in intermediate transformer activations, allowing for early detectio…

The paper demonstrates that refusal behavior in Large Language Models (LLMs) is encoded as an actionable, linearly decodable signal in intermediate transformer activations, allowing for early detectio…

The paper introduces 'abliteration,' a weight editing technique that successfully bypasses the refusal mechanism of safety-aligned Code LLMs, enabling scalable synthesis of vulnerable code from safe i…

Pramana introduces a standardized, protocol-level wire format for autonomous agent outputs, ensuring that every consequential claim is accompanied by a verifiable artifact that can be re-executed by a…

This paper investigates the privacy risk of reconstructing Personally Identifiable Information (PII) from Large Language Models (LLMs) that have undergone Supervised Finetuning (SFT), proposing a nove…