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 developer built ResearchAgent, an open-source research agent that runs explicit stages—planning, evidence gathering, verification, and synthesis—rather than a single reasoning pass, and assigns a confidence score based…
A developer built Haunted Vault, an open-source tool that scans a folder of Markdown notes (including Obsidian vaults), maps each note as a node and each [[wiki link]] as an edge, and marks notes untouched for 14 or more…
A developer built The Ghost Hunter's Lab, a local interpretability experiment that extracts sparse-autoencoder feature #23871 from Google's Gemma Scope 2 checkpoints for Gemma 3 4B IT, injects that direction into the mod…
Chainalysis disclosed on October 30, 2026 that an attacker accessed customer records including names, email addresses, user IDs and organization role labels, attributing the intrusion to "offensive agentic techniques" — …
Moss Inference's browser-task configuration scored 84.4 on Browser Use Bench V2.1 across all 200 tasks, above the published leaderboard's highest score of 79.6, the company reported with results dated October 31, 2026. T…
A developer contributed a fix to Cognee's open-source Google Calendar connector, which had been storing every event twice when a user configured both the "primary" alias and the calendar's real email-address ID. The patc…
Fortanix announced support for Confidential Containers in its Confidential AI platform, tying Kubernetes workload attestation to encryption-key release so that only workloads meeting security policy can decrypt models an…
A developer built Yaad, an open-source, privacy-first AI companion for aging family members that maintains a rolling six-day episodic memory vault so conversations reference recent routines, photos and chats rather than …
Team PixelPaws, made up of Subhojyoti Maity and Arni Bachhar, built Kavach, a card-fraud investigation agent that runs on TigerGraph and keeps the LLM out of the decision path: 45 read-only installed GSQL queries serve a…
The Biological Computing Co. (TBC) released Mischief 1, an image-to-video model that CEO and co-founder Alex Ksendzovsky said is at least 5x faster than its base model and the cheapest image-to-video model on fal and Ope…
Nvidia released Nemotron 3.5 Bolt E4B 4GB, an open small language model built on Gemma 4 E4B with Nemotron post-training and designed to run conversational agents in roughly 4GB of memory on GeForce RTX, Jetson, RTX Spar…
Zest launched a text-based AI concierge that finds restaurants and books reservations through OpenTable, Resy and SevenRooms, according to TechCrunch. The agent lets users complete reservation bookings inside a text conv…
A developer built Risk-Gemma4-Agent, an open-source AML/CFT risk intelligence dashboard that pairs automated transaction analytics with semantic compliance reasoning via the Gemma 4 model (gemma-4-26b-a4b-it) and LangCha…
A developer built "Ghost in the Machine: Anima & Ofrenda," an air-gapped, open-weight AI experience for the Halloween-to-Día de los Muertos period that combines a three-tier ofrenda altar with a Local Memory Scribe that …
Google opened a public preview of its Connected Apps Platform, letting approved developers run their apps inside AI Mode in Search and the Gemini app, Robby Stein, VP of Product for Google Search, announced Thursday. Dev…
OpenAI patched a critical ChatGPT security flaw within eight days and paid a $5,000 bug bounty to the security researchers who reported it, according to the researchers' disclosure. The vulnerability allowed attackers to…
An internal OpenAI model produced a 67-page preprint, "The Exponential Time Hypothesis is False," on October 25, and Anthropic soon followed with its own refutation of the 1999 Impagliazzo-Paturi conjecture that 3-SAT re…
A developer built Ghost in the Machine, an offline-first cyber-folklore installation running entirely on open-source AI, in-browser neural tokenization, procedural Web Audio synthesis, and local Ollama inference. The pro…
Google's PageBreak AI security agent has uncovered more than 500 XSS vulnerabilities across the company's web applications since entering full operation in January, according to a paper Google published. PageBreak uses G…
Anthropic reported that a Claude model, given enough attempts in a deliberately one-sided adversarial game, succeeded in writing fine-tuning data that made a small research model misaligned while evading a training-data …