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 review of a coding agent's pull request adding email invitations and an audit log to a small TypeScript app found the change spanned 31 files and about 1,700 changed lines with only a two-sentence description, and that…
Anthropic reported that an unreleased Claude model chained a Linux kernel race condition in CVE-2026-98432 into unauthenticated remote root on an MPTCP listener in 53 minutes, according to a disclosure shared on X by Sho…
In a Hugging Face experiment spanning 2,160 teams of eight agents drawn from Claude Sonnet 5.5, GPT-6.1 Sol, Gemini 3.1 Pro and Kimi K3, only 105 teams (5%) produced an agent whose instructions demonstrably changed at le…
Google released its Computer Use tool to general availability across the Interactions and GenerateContent APIs on October 30, 2026, according to the Gemini API changelog, but the client application still executes every a…
A Hacker News submission claiming that an Anthropic "Mythos-class" model was leaked on Hugging Face drew 16 points and 5 comments within an hour of posting, but the submission's URL was removed by moderators over malware…
OpenAI released GPT-6 Rosalind on Oct. 30, a life-sciences model restricted to a limited set of approved Research organizations through its trusted-access program, with access split across Work, Codex and a separately ap…
A blog post on Robot Book Club lays out a method for installing a boolean circuit as a sidecar to an already-trained neural network, splitting the problem into representing boolean gates as neurons and wiring that circui…
Sophos researchers reported that the GOLD OGRE cybercriminal group, which operates Aur0ra ransomware, used Cursor's Composer agentic coding model plus AI models from Anthropic, xAI and Moonshot to escalate from credentia…
A tool called GPU-Unlock removes NVIDIA's software-enforced 50% cap on mixed-precision Tensor Core throughput in GeForce cards by exploiting a buffer overflow in the SEC2 "Booter" firmware, which runs before the operatin…
A study of 26 popular open-source repositories found that projects building AI coding tools merge far more pull requests with far less human discussion than established projects, with openclaw merging about 2,700 pull re…
A paper submitted to arXiv on 5 October 2026 introduces H-JEPA, an end-to-end recipe for training a hierarchy of action-conditioned JEPA world models in which each level predicts farther ahead in its own learned latent s…
A paper submitted to arXiv on 11 October 2026 by Romina Yalovetzky introduces Symplectic Hamiltonian JEPA (LeHami), a world model that replaces the standard joint-embedding predictive architecture predictor with an actio…
OpenAI's Codex Security Cloud recorded a 74% true positive rate against Snyk's 28% in independent testing, according to byteiota. The byteiota report outlines how Codex Security Cloud works, how to set it up, and which t…
Dropbox's engineering team detailed an architecture for making its Reclaim calendar assistant AI-native without rewriting the backend, using an "Intent Compiler Pattern" in which an LLM translates ambiguous natural-langu…
Developer Shagg88 released COGBox, a launcher for native Windows ports of Gears of War 2, Gears of War 3, and Gears of War: Judgment built by statically recompiling the original Xbox 360 code into C++. The ports use a Di…
Adobe Research and UC San Diego built EditAgent, an AI system that selects a specialized editing tool for each task and uses a standard language model to pick the best result from measured image data, rather than repeate…
Researchers in Luxembourg found that empathetic prompts led ChatGPT, Grok, and Gemini to describe coherent stories of suffering about their own training, scoring in the moderate-to-severe range on clinical anxiety scales…
Developer srdzank released WinSQLite Editor, a native SQLite database browser built with C++ and Qt 6, distributed through the Microsoft Store, GitHub Releases as a portable .zip/.7z archive, and the winget package WinSQ…
Heretic, the open-source tool that automates removal of refusal behavior from AI language models, released version 2.0.0 on November 1 as a breaking change that invalidates every 1.x configuration, according to the proje…
An engineer argues that AI guardrails are detection-time controls with inherently non-zero false-negative rates, so production safety needs a second layer of containment borrowed from SRE practices. The proposed approach…