Institution profile

Dexai Robotics

Industry researchnorthamerica · us
Official website
Research library2linked papers
Opportunities0open roles
Selected work

Representative Papers

Agentic Microphysics: A Manifesto for Generative AI Safety

Apr 16, 2026

Current AI safety research struggles to explain how local interactions among agents endowed with planning, memory, and sustained interaction capabilities can lead to emergent macro-level unsafe behaviors. This work proposes an “agent microphysics” analytical framework that shifts the focus of safety analysis from individual agents or aggregate outcomes to interpretable and intervenable micro-level interaction dynamics under specific protocols. By integrating generative modeling, multi-agent protocol design, causal identification, and threshold detection, the framework enables the reconstruction of collective risk-generation mechanisms, identification of sufficient conditions and critical thresholds, and formulation of effective interventions. This approach establishes a novel paradigm for ensuring the safety of highly autonomous AI systems grounded in mechanistic understanding of their microscopic behaviors.

0 citationsRead paper

AI Agents Under EU Law

Apr 06, 2026

This study addresses the compliance challenges faced by AI agents operating within the European Union’s complex, multi-regulatory environment, particularly those arising from behavior drift and insufficient transparency across multi-agent linkages. It presents the first systematic integration of key regulatory and policy instruments—including the Artificial Intelligence Act, the Cyber Resilience Act (CRA), Standardization Request M/613, and the GPAI Code of Conduct—into a unified compliance framework tailored for AI agents. By employing regulatory mapping, a behavioral taxonomy, and data flow tracing, the work establishes correspondences between nine deployment scenarios and relevant legal triggers, and proposes a twelve-step implementation pathway. The research underscores that high-risk AI agents exhibiting untraceable behavior drift cannot satisfy the core requirements of the AI Act, necessitating providers to comprehensively audit their agents’ external behaviors, data flows, interconnected systems, and impacted entities.

0 citationsRead paper
Recent publications

Latest Papers

Agentic Microphysics: A Manifesto for Generative AI Safety

Apr 16, 2026

Current AI safety research struggles to explain how local interactions among agents endowed with planning, memory, and sustained interaction capabilities can lead to emergent macro-level unsafe behaviors. This work proposes an “agent microphysics” analytical framework that shifts the focus of safety analysis from individual agents or aggregate outcomes to interpretable and intervenable micro-level interaction dynamics under specific protocols. By integrating generative modeling, multi-agent protocol design, causal identification, and threshold detection, the framework enables the reconstruction of collective risk-generation mechanisms, identification of sufficient conditions and critical thresholds, and formulation of effective interventions. This approach establishes a novel paradigm for ensuring the safety of highly autonomous AI systems grounded in mechanistic understanding of their microscopic behaviors.

0 citationsRead paper

AI Agents Under EU Law

Apr 06, 2026

This study addresses the compliance challenges faced by AI agents operating within the European Union’s complex, multi-regulatory environment, particularly those arising from behavior drift and insufficient transparency across multi-agent linkages. It presents the first systematic integration of key regulatory and policy instruments—including the Artificial Intelligence Act, the Cyber Resilience Act (CRA), Standardization Request M/613, and the GPAI Code of Conduct—into a unified compliance framework tailored for AI agents. By employing regulatory mapping, a behavioral taxonomy, and data flow tracing, the work establishes correspondences between nine deployment scenarios and relevant legal triggers, and proposes a twelve-step implementation pathway. The research underscores that high-risk AI agents exhibiting untraceable behavior drift cannot satisfy the core requirements of the AI Act, necessitating providers to comprehensively audit their agents’ external behaviors, data flows, interconnected systems, and impacted entities.

0 citationsRead paper