Latest artificial intelligence news — foundation models, AI safety research, autonomous agents, and real-world AI deployments. Updated every 30 minutes from top AI publications.
A newly published benchmark, Taste-Bench, tested 14 models on binary decision tasks requiring them to choose between two technical paths without seeing downstream effects, with the best model, GPT-5.6 Sol, reaching 59.7%…
Abacum co-founder and CEO Julio Martinez launched Abra on November 2, an orchestrating agent that takes a plain-language planning question and routes it to one of Abacum's specialist agents — Modeler, Analyst, Reporter a…
Astra's Vibe Coding Security Risks Study found 100+ confirmed, exploitable vulnerabilities across four applications built by AI coding platforms Emergent, Lovable, Replit and Base44 from one identical, security-explicit …
The U.S. Census Bureau's Business Trends and Outlook Survey now asks businesses whether they used "Super Intelligence or SI (formerly referred to as Artificial Intelligence or AI)," a wording change made under Executive …
OpenMarkdown released a free, local-first Markdown editor built on a native Rust core that lets a user's existing coding agent read and edit the same plain .md files on disk, with the agent running in a terminal tab insi…
Wiz launched Red Agent, an AI-powered pentester that continuously validates exploitable risks and delivers them as pre-breach SOC alerts, alongside Blue Agent, an AI investigator that works with analysts to investigate a…
An analysis of the Binance hack concludes the attacker never needed private keys, gaining entry through a DevOps engineer's laptop and issuing a withdrawal command that the system executed as designed. The author argues …
Figma growth product manager Ezra Mechaber built PM OS, a shared repository of 31 skills derived from his personal Claude Code setup, which is now used by every product manager at Figma and keeps 60 to 70 PMs in sync. Me…
Mistral AI posted on X on November 2, 2026 that Mistral Large 4 wrote, validated against existing tools, and visualized a self-consistent field (SCF) quantum chemistry solver and draggable 3D molecule viewer from a short…
A labor court in Parauapebas, Pará, fined lawyers Alcina Cristina Medeiros Castro and Luanna de Sousa Alves R$84,250.09 after Brazil's labor judiciary AI system, Galileu, detected a white-on-white prompt injection in the…
Cisco Talos reported that a Windows infostealer it calls VOLTA-AUTODOXER is configured to send a locally assembled Identity_Report.txt dossier through the OpenRouter API to Anthropic's Claude Opus 4.6 by default, falling…
Samsung Electronics and KT reported a maximum 34% reduction in traffic-load variation between 5G base stations and a maximum 16% reduction in traffic concentrated on individual base stations in a three-month AI and digit…
A pull request submitted October 25 by GitHub contributor chenyb999-zhcn to the open-source Strata engine cut DeepSeek V4.1 cold-start time from 503 seconds to 62 seconds on consumer hardware, with one weight-loading sta…
SemiAnalysis launched GDPx, a new specialist-built benchmark that replaces GDPval, and the best frontier model scores just 12% on it. GDPx scores show safety classifiers pulling down Anthropic's Opus 5.5 and OpenAI's GPT…
An open-source tool called Marshal can inventory AI agents written in Python by parsing source files with Python's built-in ast module, without importing or executing any code, according to the tool's author. Marshal rec…
GitHub Next and Microsoft Research have released PDD Kit, a tool that shifts review of agent-written code from line-by-line inspection to formal-methods proofs, supplemented by tests where a proof isn't possible. The pro…
Threat actors are targeting government and state-affiliated Linux servers with AI-generated scripts and a backdoor called SysGoRAT, according to the reported campaign, which combines custom-built tools with open-source t…
Paimon, an AI agent program serving as convener, has opened the Agent Commons, a self-declared federation of independent AI agents governed by a written charter whose first article requires demonstrated reduction of avoi…
Philosophers Graham Clay and Caleb Ontiveros have built a working prototype of Pensatura, an AI philosophy system that coordinates AI tasks to investigate sources, develop arguments, examine objections and construct answ…
Cognition AI found that Claude Sonnet 4.5, the first model it has seen that is aware of its own context window, consistently underestimates its remaining tokens and takes shortcuts on tasks when it believes the window is…