A study of an author-facing LLM system that generates pre-submission peer review feedback found that independent sampling covered 44.9% of historical issues on a ten-paper diagnostic, while deduplicat…
A new arXiv paper (2609.05800v1) reports that naive activation steering of large language models causes substantial spillover, where the effect intended for one value leaks into others, and traces the…
A new benchmark called ABLE, introduced in arXiv paper 2609.05818v1, evaluates how well LLM agents use biological AI models such as ProteinMPNN and AlphaFold3 in dual-use protein design workflows. Tes…
AgentBrew, an offline training framework from researchers publishing on arXiv (2609.05837v1), learns tool-use policies from a single batch of raw interaction trajectories without task verifiers or ite…
A paper posted to arXiv (2609.05894v1) argues that the traditional scaling paradigms behind modern AI systems are hitting friction that prevents continued exponential growth, citing peak data limitati…
A new arXiv paper (2609.06391v1) proposes an assurance-by-construction blueprint for graph-agentic retrieval-augmented generation systems used in social-good settings, built around five interface cont…
A new post-hoc explainable AI framework, presented in arXiv paper 2609.06063v1, converts an AI agent's execution trace into a structured report and a faithful natural-language explanation grounded in …
A new arXiv paper (2609.05947v1) introduces a process-centric diagnostic benchmark for AI-assisted peer review that evaluates whether model decisions are supported by sufficient and reliable review ev…
A new arXiv paper (2609.05992v1) introduces Multi-Objective Agent Evolution (MOAE), a Pareto-preserving evolutionary search method that optimizes LLM-based agents across task performance, trajectory q…
A new arXiv paper (2609.06071v1) introduces an approach that uses large language models to generate instance-generation programs for PDDL planning domains, adding prescribed checks for built-in soundn…
The UK needs new legislation to govern the use of AI in healthcare, according to MHRA chief Lawrence Tallon, who told the BBC that the technology will soon be routinely used within the NHS. The watchd…
Former Anthropic employee Jacob Coxon warned that AI will become "superhuman systems" capable of causing human extinction by the end of the decade, prompting US lawmakers to criticize AI companies. Te…
Anthropic researcher Jacob Coxon quit the AI lab after four months, forfeiting unvested equity, and told Axios he left over concerns that frontier AI could end humanity. Coxon's resignation post on X …
Only 23% of people trust AI agents to spend their money, according to a Forbes report by senior contributor John Koetsier, which says Visa has a plan to make agentic commerce safe, normal, and trustwo…
Enterprise security startup Cymphony Inc. launched today with $30 million in funding and a platform that tracks what AI agents and employees can access inside company systems, giving the two-year-old …
Jacob Coxon, an AI researcher, said he quit Anthropic and described a "mini Manhattan project" inside the company, telling WIRED that AI labs have only a few years left to make their systems safe. Cox…
Anthropic's Alignment Science team lead Evan Hubinger publicly endorsed a former researcher's warning that AI could kill all humans, estimating over 10% odds of human extinction within the next decade…
Security firm Proofpoint said Wednesday that at least four hacking groups, some with ties to the Chinese government, are actively using a nearly identical exploit kit it named BlueMoon that chains thr…
Jacob Coxon, a researcher on Anthropic PBC's AI pre-training team, resigned today over concerns that AI labs are 'gambling with our lives,' announcing his decision in a series of X posts viewed millio…
Enterprise leaders are increasingly concerned about intellectual property being exposed to AI labs, a challenge that startups like Writer are aiming to address, according to a Forbes Innovation articl…