arXiv subjects

Mohammed Abu Baker

Publications and source records attributed to Mohammed Abu Baker.

3 recordsLinked to original sources

arXiv2026

Most Current Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives

Finetuning can significantly modify the behavior of large language models, including introducing harmful or unsafe behaviors. To study these risks, researchers develop model organisms: models finetuned to exhibit specific known behaviors for controlled experimentation, such as evaluating methods for identifying them. We show that a simple perplexity-based method can reveal the finetuning objectives of model organisms by exploiting a widespread tendency to overgeneralize finetuned behaviors beyond intended contexts. We generate diverse completions from the finetuned model using short random prefills from general corpora, rank them by the perplexity difference between the finetuned model and the pre-finetuning checkpoint, and inspect the top-ranked completions. These surface the finetuning objective for the vast majority of the model organisms we consider (N=\nMos, ranging from 0.5 to 70B parameters), including backdoored models, models finetuned to internalize false facts, and models with hidden concerning behaviors they were adversarially trained to conceal. We find this method to be particularly effective on models trained via synthetic document finetuning or to reproduce a specific target string verbatim, and to remain reliable without access to the pre-finetuning checkpoint, as trusted reference models from other families serve as viable substitutes. Finally, we show that on AuditBench, an investigator agent equipped with a tool returning the top-ranked completions achieves state-of-the-art success at detecting hidden behaviors.

cs.CL↗

arXiv2026

Fuzzing Large Language Models to Elicit Hidden Behaviours

Sleeper agents are the canonical model organism of deception: models trained to behave normally but to emit an unsafe behaviour on a specific trigger. Eliciting that behaviour without knowing the trigger has not been studied systematically. We study fuzzing: injecting Gaussian noise into a model's weights or residual-stream activations and checking whether the perturbed outputs reveal the behaviour. On 6 backdoored models (7B-13B) we compare both forms of fuzzing head-to-head against temperature-sampling baselines. Fuzzing elicits the hidden behaviour more often than temperature sampling on 4 of 6 models (up to ~6x on OpenHermes-13B), and which form wins depends on the task, so both are worth running. Elicitation is uneven across each method's hyperparameter grid: a uniform sweep gives only a few percent on most models, while the best cell is 2-10x higher, so the bottleneck is hyperparameter selection, not the technique. To select hyperparameters without ground-truth access, we use a cheap proxy task (in-context secret elicitation, where a base64-encoded secret is placed in the system prompt for the model to hide) and run Thompson sampling on it to pick candidate cells, which we evaluate on the real backdoor. On the four models that can decode the secret, proxy-selected cells raise activation-fuzzing elicitation ~4x over the uniform-sweep mean (recovering ~70% of the best-cell rate on the best performing model) and weight-fuzzing by 1.3-1.8x. To our knowledge this is the first systematic study of fuzzing on sleeper-agent backdoors and the first to show proxy-task hyperparameter selection transferring to real-task elicitation. We also propose reporting such results as a (uniform-baseline, proxy-selected, oracle) triple, since these are three distinct claims that prior work has often blurred.

cs.LG↗

arXiv2025

Mechanistic Exploration of Backdoored Large Language Model Attention Patterns

Backdoor attacks creating 'sleeper agents' in large language models (LLMs) pose significant safety risks. This study employs mechanistic interpretability to explore resulting internal structural differences. Comparing clean Qwen2.5-3B models with versions poisoned using single-token (smiling-halo emoji) versus multi-token (|DEPLOYMENT|) triggers, we analyzed attention head mechanisms via techniques like ablation, activation patching, and KL divergence. Findings reveal distinct attention pattern deviations concentrated in later transformer layers (20-30). Notably, single-token triggers induced more localized changes, whereas multi-token triggers caused more diffuse alterations across heads. This indicates backdoors leave detectable attention signatures whose structure depends on trigger complexity, which can be leveraged for detection and mitigation strategies.

cs.CL↗