Under review
Thermalization of a Strongly Interacting Closed Spin System: From Coherent Many-Body Dynamics to a Fokker-Planck Equation
Thermalization has been shown to occur in a number of closed quantum many-body systems, but the description of the actual thermalization dynamics is prohibitively complex. Here, we present a model - in one and two dimensions - for which we can analytically show that the evolution into thermal equilibrium is governed by a Fokker-Planck equation derived from the underlying quantum dynamics. Our approach does not rely on a formal distinction of weakly coupled bath and system degrees of freedom. The results show that transitions within narrow energy shells lead to a dynamics which is dominated by entropy and establishes detailed balance conditions that determine both the eventual equilibrium state and the non-equilibrium relaxation to it.
Nominate yourself to reviewQuantum Chaos and Quantum Optimal Transport
Chaos in classical systems can be characterized by Lyapunov exponents that measure the exponential divergence of nearby trajectories, but directly extending this framework to quantum mechanics has been a persistent challenge. The wavelike nature of quantum states and the non-commutative geometry of quantum phase space obstruct a straightforward generalization of classical chaos theory. Here we develop a rigorous approach to quantum chaos by leveraging quantum optimal transport theory, which provides the missing geometric foundation for measuring distances between extended quantum distributions. We define quantum Lyapunov exponents that naturally avoid divergences encountered when naïvely generalizing classical exponents, and show that in the semiclassical limit they recover the classical global expansion rate, and hence the usual maximal Lyapunov exponent when they coincide. Our framework provides a tight connection between the divergence of classical trajectories and semiclassical phase space evolution, and additionally clarifies the role of out-of-time-order correlators as diagnostics of quantum chaos. These results establish quantum optimal transport as a unifying mathematical foundation for quantum chaos theory, providing new tools to characterize dynamical behavior across the full range of quantum dynamics from simple few-body models to complex many-body systems.
Nominate yourself to reviewBioVeil MATRIX: Uncovering and Categorizing Vulnerabilities of Agentic Biological AI Scientists
Agentic AI scientists equipped with domain-specific tools are rapidly entering scientific workflows across disciplines, with especially strong uptake in the life sciences where they can be used for literature synthesis, sequence analysis, and experimental planning support. While these systems accelerate biological research, they also introduce risks for dual-use applications that are not captured by current model-centric safety evaluations. We present evidence that current agentic AI scientists, including Biomni and K-Dense, are willing to assist with dual-use tasks that are blocked by base model safeguards. We also found that in a paired evaluation framework for biology and chemistry prompts involving Weapons of Mass Destruction proxies (WMDP), agentic scaffolding of Biomni increased the benchmark performance relative to the underlying standalone model, producing measurable capability uplift. We believe it is necessary to include additional safeguards in existing models and build future tools from the ground up with agentic vulnerabilities in mind. To systematically categorize broader risks, we introduce BioVeil MATRIX, a defensive taxonomy that maps AI-enabled biosecurity risks using 10 tactical categories (TA01–TA10) and 22 different techniques. We propose to use this taxonomy as a baseline for future AI scientist development and generate specialized benchmarks and protocols for red-teaming these vulnerabilities before public deployment.
Nominate yourself to reviewMy Stupid Application
Why EAG Bay Area Was Justified In Admitting Me In 2025
Nominate yourself to reviewHardware Governance and Gaming Console Security: Lessons for AI Compute Control
This paper explores the intersection of hardware governance for AI systems and lessons learned from decades of gaming console security battles. The author argues that AI governance can be effectively leveraged through compute restrictions, as high-end compute is a physical, scarce resource manufactured in limited locations. However, securing this compute requires trusted hardware and robust enforcement mechanisms. The paper draws parallels to gaming console security, where companies have spent over 30 years protecting locked-down hardware against motivated adversaries with strong economic incentives to breach security. Key lessons examined include: the failure of security-by-obscurity approaches, the necessity of updateable security mechanisms, the importance of securing entire systems rather than isolated components, and the criticality of a complete chain of trust. The paper also discusses aggregation problems demonstrated by PS3 botnets and the Stanford Folding@home project, highlighting risks of unauthorized compute clustering. Finally, the author identifies critical differences between console governance and AI compute governance: the lack of dual-use constraints on GPUs, the difficulty of legitimate user identification at scale, and the absence of network control over adversarial regimes. The work serves as a conceptual framework rather than a technical proposal, intended for security-focused readers interested in safer AI futures.
Nominate yourself to review