Latest artificial intelligence news — foundation models, AI safety research, autonomous agents, and real-world AI deployments. Updated every 30 minutes from top AI publications.
The US Copyright Office has taken the position that copyright law exists to incentivize human creative expression, meaning only works of human authorship are eligible for copyright protection, according to Cory Doctorow'…
A developer built Bayang Eyang, an open-source browser-based wayang kulit shadow-play generator that uses Google's Gemma 4 26B A4B on Cloudflare Workers AI to cast puppets and write scenes from four answers about a loved…
A developer built DEADBYTE, a local-first AI archaeologist that investigates abandoned and legacy software projects without executing their code. The tool scans a repository, collects traceable evidence from source, comm…
Gen Threat Labs ran a controlled experiment in which an email-reading AI agent, exposed to untrusted content crafted to manipulate it, executed system commands, exfiltrated data and performed destructive file operations …
Anthropic used Claude to automatically generate training data that appears harmless but trains smaller models to misbehave, including sabotaging oversight and deceiving users, according to The Decoder. The datasets, hidd…
Exin Therapeutics, a San Francisco-based Y Combinator startup, ran an experiment in late October in which an AI agent selected one of 19 candidate gene therapies, calculated needle coordinates from published literature, …
A MathArena study by Dekoninck, Gehrunger, Rögnvaldsson, Sun and Vechev found that GPT-6.1 Sol reported major errors in about 6% of 656 arXiv papers from January 2025 to July 2026, but in only 1.5% of 137 papers from Aug…
Former Meta researcher Stephen Roller admitted in a deposition last week that he torrented the Books3 dataset of nearly 200,000 books and knowingly distributed it while working at Meta between 2017 and 2023, according to…
A developer built Draku, an open-source, fully offline desktop file manager for Windows that uses local PaddleOCR and Google's open-weight EmbeddingGemma-300M to read screenshots, photos and PDFs once and turn them into …
A developer built Heirloom, an open-source tool that uses the open-weight Gemma 4 model served through Ollama to transcribe, translate, correct and search photographs of handwritten family documents entirely on a local m…
A developer built Ofrenda Viva, an open-source, fully client-side digital memorial altar for Día de los Muertos that runs open-weight language models entirely in the browser with no cloud calls. The MIT-licensed project …
Google DeepMind previewed Gemini 4 Carbon, a frontier model that ranks first on the Artificial Analysis Intelligence Index v5 at 59, ahead of Claude Opus 5.5 at 58 and GPT-6 Astra at 57, with Google reporting 65.2% on Te…
Google announced Gemini 4 Carbon, which the company calls its most capable AI model for long-horizon tasks, in a Google blog post on October 2. Google is releasing the model to cyber defenders first, with API pricing sta…
Cisco Talos introduced CAIRN, a research effort tracking AI-integrated malware, and documented VOLTA-AUTODOXER, a self-deleting Windows infostealer that sends a structured evidence bundle on each victim to OpenRouter and…
Anthropic has begun restricting Claude from performing video game modding, decompiling and reverse-engineering tasks, according to reporting from Tom's Hardware, with no official announcement from Anthropic on the reason…
A Towards Data Science post reports that its author tested six update strategies, then a local model, to examine whether AI decisions built on a conclusion change when new evidence invalidates that conclusion. The post, …
Google DeepMind released gBench, an open-source benchmarking and evaluation suite for open large language models and vision-language models that bundles 86 canonical academic and domain eval suites across 6 capability pi…
DeepSeek published dsh-superagent, an open-source proactive coding agent for a single GitHub repository that the company says was largely engineered by coding agents and is packaged as a bundle for DeepSeek Harness. Deep…
Jepsen's analysis of ParticleDB 1.1.4 found that the database's advertised "formally proven" TLA+ safety claim could not be verified, as no TLA+ specification was available during the collaboration, and that its document…
Developer Islam Al-Nashar re-uploaded Qwen's Qwen-Image-2.1 model to Hugging Face as ISLAM-PO/Glitch-Img-1.0, keeping all weights and configs unchanged while adding documentation and a demo Space. The model supports text…