Research
A paper is listed when one of its authors is affiliated with the lab. Data: OpenAlex.
Scaling Laws for Safety ProbesAnthropic · arXiv (Cornell University)Oct 22A Bird's-Eye View of Iterative Reward DesignAnthropic · arXiv (Cornell University)Oct 3Borges: A Low-Latency Distributed Shared Log on a CXL Memory/SSD HybridAnthropic · ACM SIGOPS Symposium on Operating Systems Principles (SOSP)Sep 28When Do Models Admit They Are Wrong? Failure Disclosure Is Unstable Under Reinforcement LearningAnthropic · arXiv (Cornell University)Sep 27Minmax-Regret k-Sink Location on a Dynamic Tree Network with uniform capacitiesAnthropic · Discrete Applied Mathematics4 citationsAug 27Mind Viruses: Self-Propagating Ideas in Multi-Agent LLM SystemsAnthropic · arXiv (Cornell University)Aug 10A Shared Subcircuit Lets LLMs Count Down Across TasksAnthropic · arXiv (Cornell University)Jul 14Masked attribution-based probing of strategies as a computational framework to align human, non-human primate, and model explanationsAnthropic · Communications PsychologyJul 13Covert Attacks on Machine Learning Training in Passively Secure MPCAnthropic · IEEE European Symposium on Security and PrivacyJul 1The Value Axis: Language Models Encode Whether They're on the Right TrackAnthropic · arXiv (Cornell University)Jun 15A Type Theory of Sense: Witnessed Choice in Stratified Semantic Spaces — paper and supplementary dataAnthropic · arXiv (Cornell University)Jun 12Diffuse AI Control on Fuzzy TasksAnthropic · arXiv (Cornell University)Jun 8Why Larger Models Learn More: Effects of Capacity, Interference, and Rare-Task RetentionAnthropic · arXiv (Cornell University)May 28From Simulation to Enaction: Post-trained language models recognize and react to their own generationsAnthropic · arXiv (Cornell University)May 25Faithfulness as Information Flow: Evaluating and Training Faithful Chain-of-Thought ReasoningAnthropic · arXiv (Cornell University)May 22Incalmo: an Autonomous Llm-Assisted System for Red Teaming Multi-Host NetworksAnthropic · Proceedings - IEEE Symposium on Security and Privacy/Proceedings of the ... IEEE Symposium on Security and Privacy7 citationsMay 18SLEIGHT-Bench: A Benchmark of Evasion Attacks Against Agent MonitorsAnthropic · arXiv (Cornell University)May 15ExploitGym: Can AI Agents Turn Security Vulnerabilities into Real Attacks?Alphabet, Anthropic, OpenAI · arXiv (Cornell University)3 citationsMay 11Removing Sandbagging in LLMs by Training with Weak SupervisionAnthropic · arXiv (Cornell University)Apr 23Language models transmit behavioural traits through hidden signals in dataAnthropic · Nature9 citationsApr 15Faster Game Solving via Hyperparameter SchedulesAnthropic · Proceedings of the AAAI Conference on Artificial IntelligenceMar 14Triton-Sanitizer: A Fast and Device-Agnostic Memory Sanitizer for Triton with Rich Diagnostic ContextAlphabet, Anthropic, Meta, OpenAI · International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS)1 citationMar 10Curation Leaks: Membership Inference Attacks against Data Curation for Machine LearningAnthropic · arXiv (Cornell University)Feb 28Haiku to Opus in Just 10 bits: LLMs Unlock Large Compression GainsAnthropic · arXiv (Cornell University)Feb 9Taming Multidimensional Analytics Costs: An Empirical Study of HLL and KLL Sketches in Cloud Data WarehousesAnthropic · IEEE SoftwareFeb 3