Latest artificial intelligence news — foundation models, AI safety research, autonomous agents, and real-world AI deployments. Updated every 30 minutes from top AI publications.
Anthropic disclosed a zero-click Linux kernel exploit, tracked as CVE-2026-98432, that lets an unauthenticated attacker execute arbitrary commands as root on servers running Multipath TCP (MPTCP) listeners, using an unre…
Ramen launched Aura Autopilot, an open-source autonomous game playtesting agent for Unreal Engine, Unity and Godot, publishing it on GitHub alongside a whitepaper benchmarking it against rival game development agents. Ra…
Exin Therapeutics, a Y Combinator startup building AI agents for brain drug discovery, says one of its agents selected a gene therapy candidate and performed the injection stage of brain surgery on a mouse with epilepsy,…
GitHub unveiled a redesigned Copilot app and a new Copilot Projects capability at its annual GitHub Universe conference, letting teams assign broader goals spanning multiple repositories and distribute work across multip…
A developer built Spookify, a lightweight Gradio web app that turns mundane daily activities into two-paragraph horror stories using the open-weight Llama 3 model (llama3-8b-8192) served through the Groq API for fast inf…
A practitioner's account of a RAG evaluation workflow reports that on a 200-question run, reciprocal rank fusion at k=5 reached a hit-rate of 0.72, versus 0.61 for dense retrieval at k=5 and 0.54 for BM25 at k=5. The aut…
Cloudera Inc. launched Cloudera Anywhere Cloud, a hybrid data and AI platform spanning public clouds and on-premises environments, with the company arguing that role-based access control rather than model choice determin…
Prosper Global, formerly Mercy Corps, built a multi-agent research engine called VERA with Cloudera that cut desk research for an agricultural report in Sudan from five or six days to two, saving about $1,500, according …
Researchers proposed Formula-Driven Supervised Metric Learning (FDSML), a pre-training method that replaces the classification-on-discretized-labels approach used in Formula-Driven Supervised Learning (FDSL) with metric …
An independent reproduction of Zhang Chong's December 2025 paper "Attention Is Not What You Need" found a 22.6% performance gap versus the paper's claim that its "Grassmann Flows" architecture lands within 10-15% of size…
Researchers introduced Holo4D, a method that uses holistic 4D reconstruction as geometric control for video diffusion, addressing the problem of synthesizing a monocular video from a new camera trajectory while preservin…
Vibewright, a desktop app for Windows 10/11 (64-bit) and macOS 11 Big Sur or later, turns a one-line project idea into an approved roadmap and then builds it with Claude Code running in an embedded terminal inside the ap…
Zocdoc announced it has opened its platform to AI agents, letting agents integrate with Zocdoc's infrastructure to find providers, verify insurance plans, and book and add appointments to patients' calendars. Zocdoc said…
Trida AI published a position arguing that prompt injection cannot be reliably prevented by prompting models to distinguish instructions from data, and that defenses must instead be architectural. The company recommends …
AI agents drafted more than 40% of all pull requests on GitHub in September 2026 — 110 million in one month — up from 17% six months earlier, according to GitHub's Octoverse 2026 report released Tuesday. GitHub counted 3…
Anthropic says an unreleased model autonomously discovered and chained browser bugs into a sandbox escape, producing working exploits that led to 62 patched Chrome CVEs, with little human involvement. Epoch found the mod…
A project called rr-decomp has produced a byte-matching C decompilation of the Japanese PlayStation 1 release of Ridge Racer (SLPS-00001), implementing all 408 game functions in C compiled with GCC 2.5.7 to yield an exec…
A 2021 Nature paper claiming a reinforcement learning agent placed chip macros in under six hours with better power, timing and area than human engineers has not replicated, with Cheng and Kahng at UC San Diego unable to…
Setterly's AI sales agent booked 105 qualified home surveys in its first 30 days for UK electric-heating company Fischer Future Heat, which routes every enquiry still unbooked after seven days to the agent, according to …
A developer reports that AI coding agents still fail to account for security unless explicitly prompted, producing code that regularly violates OWASP Top 10 basics. A study by Bappy et al. (2026) found that only two of 1…