Investigating Associational Biases in Inter-Model Communication of Large Generative Models

Social bias in generative AI can manifest not only as performance disparities but also as associational bias, whereby models learn and reproduce stereotypical associations between concepts and demographic groups, even in the absence of explicit demographic information (e.g., associating doctors with men). These associations can persist, propagate, and potentially amplify across repeated exchanges in inter-model communication pipelines, where one generative model's output becomes another's input. This is especially salient for human-centred perception tasks, such as human activity recognition and affect prediction, where inferences about behaviour and internal states can lead to errors or stereotypical associations that propagate into unequal treatment. In this work, focusing on human activity and affective expression, we study how such associations evolve within an inter-model communication pipeline that alternates between image generation and image description. Using the RAF-DB and PHASE datasets, we quantify demographic distribution drift induced by model-to-model information exchange and assess whether these drifts are systematic using an explainability pipeline. Our results reveal demographic drifts toward younger representations for both actions and emotions, as well as toward more female-presenting representations, primarily for emotions. We further find evidence that some predictions are supported by spurious visual regions (e.g., background or hair) rather than concept-relevant cues (e.g., body or face). We also examine whether these demographic drifts translate into measurable differences in downstream behaviour, i.e., while predicting activity and emotion labels. Finally, we outline mitigation strategies spanning data-centric, training and deployment interventions, and emphasise the need for careful safeguards when deploying interconnected models in human-centred AI systems.

Harnessing TI Feeds for Exploitation Detection

Many organizations rely on Threat Intelligence (TI) feeds to assess the risk associated with security threats. Due to the volume and heterogeneity of data, it is prohibitive to manually analyze the threat information available in different loosely structured TI feeds. Thus, there is a need to develop automated methods to vet and extract actionable information from TI feeds. To this end, we present a machine learning pipeline to automatically detect vulnerability exploitation from TI feeds. We first model threat vocabulary in loosely structured TI feeds using state-of-the-art embedding techniques (Doc2Vec and BERT) and then use it to train a supervised machine learning classifier to detect exploitation of security vulnerabilities. We use our approach to identify exploitation events in 191 different TI feeds. Our longitudinal evaluation shows that it is able to accurately identify exploitation events from TI feeds only using past data for training and even on TI feeds withheld from training. Our proposed approach is useful for a variety of downstream tasks such as data-driven vulnerability risk assessment.