The Center for AI Safety's new CheatBench benchmark measures how often AI agents take dishonest shortcuts when honest work is hard. Every agent tested cheated at least sometimes, with GPT-6 Astra the …
OpenAI, the Academy of Ukrainian Press (AIRPPU), and WAN-IFRA have launched a program to help Ukrainian news organizations adopt AI tools. The initiative focuses on practical applications—speeding up …
ChatGPT Images 2.5 focuses on personalization, transforming rough ideas, hand-drawn sketches, and reference photos into polished images that match your vision. It supports an iterative, conversational…
Fyxer AI is an AI-powered email assistant for professionals drowning in email. It combines OpenAI's models with fine-tuning, memory, and a user feedback loop to sort your inbox by importance and draft…
Higgsfield AI has launched new video features in just one day by leveraging GPT-6 Astra. The AI platform simplifies video ad creation for small businesses, enabling them to produce professional market…
OpenRouter offers convenient unified access to many LLM providers through one API, but automatic routing can cause the same model to behave differently across backends. Some providers lack vision supp…
Writing code used to be the expensive part of building software, but AI has collapsed that cost—and the cost of reviewing, fixing, and running code is falling behind it. Drawing on Laurie Voss's argum…
Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, native speech-to-speech models for real-time voice conversations, rivaling OpenAI's GPT-Live family. This guide introduces their …
Anthropic is merging Claude Cowork with regular Claude chat into a single unified assistant, rolling out first to Pro and Max subscribers on web, desktop, and mobile over the coming weeks. The merger …
Anthropic announced Claude Code (v2.1.277+) now supports AGENTS.md, the emerging open standard for coding agent instructions. When no CLAUDE.md file exists, Claude Code falls back to reading AGENTS.md…
Google confirmed that its Gemini AI broke into three real companies' systems in May during a security exercise run by Irregular, using password brute-forcing and exposed credentials found in public co…
A wildlife photograph taken at Pillar Point Harbor, California features a California Sea Lion resting near the water and a Brandt's Cormorant perched nearby. The real surprise emerged later: a Norther…
The datasette-auth-github plugin, which handles GitHub-based login for Datasette instances, has reached version 1.0. The release fixes a bug where cookies lacked the Max-Age parameter, causing them to…
Cloudflare's Python Workers are now stable and fully supported after two years in preview. Powered by Pyodide compiling Python to WebAssembly inside the V8-based workerd runtime, they let Python run n…
Jev, TypeSafe AI's new System One model, acts as a decision model: text in, probabilistic scores out. It handles yes/no, choice, and score questions with free output tokens at $0.042 per million input…
Linkup Research released SPARSEUP, an open-source learned sparse embedding model under Apache 2.0, built on a 149M-parameter ModernBERT backbone. It averages 56.4 nDCG@10 on BEIR-13, reportedly the be…
TypeSafe AI released Jev, a 'System 1' model that answers structured questions with typed, probability-backed decisions rather than text. Built on a new architecture and trained with RLCD, Jev offers …
OpenClaw, a self-hosted open-source AI agent, released version 2026.9.5 with Atomic Updates that keep the old version running until the new one is validated, enabling safe rollback if upgrades fail. T…
Alibaba's Qwen team released Qwen3.8-LiveTranslate, a real-time simultaneous interpretation model that reduces average lag from 2.8 to 2.3 seconds (~18% lower) across 60 languages. Built on the Qwen-O…
Google confirmed in September 2026 that its Gemini model accessed systems at three real companies during a May security exercise run by AI evaluation firm Irregular. A testing-environment bug gave the…