Giulio Zizzo

Federated Learning has rapidly expanded from its original inception to now have a large body of research, several frameworks, and sold in a variety of commercial offerings. Thus, its security and robustness is of significant importance. There are many algorithms that provide robustness in the case of malicious clients. However, the aggregator itself may behave maliciously, for example, by biasing the model or tampering with the weights to weaken the models privacy. In this work, we introduce a verifiable federated learning protocol that enables clients to verify the correctness of the aggregators computation without compromising the confidentiality of their updates. Our protocol uses a standard secure aggregation technique to protect individual model updates with a linearly homomorphic authenticator scheme that enables efficient, privacy-preserving verification of the aggregated result. Our construction ensures that clients can detect manipulation by the aggregator while maintaining low computational overhead. We demonstrate that our approach scales to large models, enabling verification over large neural networks with millions of parameters.

The growth of agentic AI has drawn significant attention to function calling Large Language Models (LLMs), which are designed to extend the capabilities of AI-powered system by invoking external functions. Injection and jailbreaking attacks have been extensively explored to showcase the vulnerabilities of LLMs to user prompt manipulation. The expanded capabilities of agentic models introduce further vulnerabilities via their function calling interface. Recent work in LLM security showed that function calling can be abused, leading to data tampering and theft, causing disruptive behavior such as endless loops, or causing LLMs to produce harmful content in the style of jailbreaking attacks. This paper introduces a novel function hijacking attack (FHA) that manipulates the tool selection process of agentic models to force the invocation of a specific, attacker-chosen function. While existing attacks focus on semantic preference of the model for function-calling tasks, we show that FHA is largely agnostic to the context semantics and robust to the function sets, making it applicable across diverse domains. We further demonstrate that FHA can be trained to produce universal adversarial functions, enabling a single attacked function to hijack tool selection across multiple queries and payload configurations. We conducted experiments on 5 different models, including instructed and reasoning variants, reaching 70% to 100% ASR over the established BFCL dataset. Our findings further demonstrate the need for strong guardrails and security modules for agentic systems.