Public articles linked to the same research event.
arXiv Pincer is a resource-layer defense whose core is a digital twin, an isolated-context model that automatically learns and enforces dynamic user-specific least-privilege policies and acts as the user's proxy for an agent's permission requests; the authors propose a user-centric dataset built from a multi-day user-agent interaction transcript to emulate the learning phase, and their evaluation shows Pincer performs strongly on both security and utility against baselines including variants of LLM judges and adaptations of Conseca (HotOS '25), with significant security improvement on some attack types.
Pincer is a resource-layer defense whose core is a digital twin, an isolated-context model that automatically learns and enforces dynamic user-specific least-privilege policies and acts as the user's proxy for an agent's permission requests; the authors propose a user-centric dataset built from a multi-day user-agent interaction transcript to emulate the learning phase, and their evaluation shows Pincer performs strongly on both security and utility against baselines including variants of LLM judges and adaptations of Conseca (HotOS '25), with significant security improvement on some attack types.
Pincer is a resource-layer defense whose core is a digital twin, an isolated-context model that automatically learns and enforces dynamic user-specific least-privilege policies and acts as the user's proxy for an agent's permission requests; the authors propose a user-centric dataset built from a multi-day user-agent interaction transcript to emulate the learning phase, and their evaluation shows Pincer performs strongly on both security and utility against baselines including variants of LLM judges and adaptations of Conseca (HotOS '25), with significant security improvement on some attack types.
Pincer is a resource-layer defense whose core is a digital twin, an isolated-context model that automatically learns and enforces dynamic user-specific least-privilege policies and acts as the user's proxy for an agent's permission requests; the authors propose a user-centric dataset built from a multi-day user-agent interaction transcript to emulate the learning phase, and their evaluation shows Pincer performs strongly on both security and utility against baselines including variants of LLM judges and adaptations of Conseca (HotOS '25), with significant security improvement on some attack types.