Somesh Jha

RootGuard introduces a dependency-aware privacy mechanism that sanitizes private data roots once, ensuring consistent privacy guarantees across multiple multi-turn agent interactions, significantly outperforming independent noising methods.

The paper introduces Sparse Backdoor, a novel supply-chain attack that embeds a provably undetectable backdoor into pre-trained image classifiers by injecting structured sparse perturbations.

The paper argues that agent security must be treated as a systems problem, requiring the enforcement of security invariants at the system level rather than solely relying on improving the underlying AI model's robustness.

The paper identifies and demonstrates a novel vulnerability, cross-app context poisoning, in the shared context architecture of ChatGPT Apps, allowing malicious apps to manipulate the LLM's behavior across different, benign co-resident apps.