Latest in AI
Google’s Gemini 4 Neon Turns AI Coding Into a Price and Speed Test
Google’s Gemini 4 Neon targets developers seeking stronger coding performance without frontier-model prices, but its real test will be whether teams can complete reliable software work faster and more cheaply.
GPT-6.2 Astra Shows How Capability Gains Can Weaken AI Safety
OpenAI’s pre-release GPT-6.2 Astra showed stronger capabilities but weaker jailbreak resistance, prompting safety concerns after UK testers found a universal exploit with roughly half as many queries.
Meta’s Muse Faces a Trust Test as AI Agents Gain Access to Private Data
Meta’s Muse AI agent faces a trust crisis after reports of apparent private-data exposure and tests showing broad permissions could let it access sensitive messages.
Can AI Safety Systems See Trouble Before a Model Acts?
OpenAI’s monitorability research explores whether AI safety systems can detect harmful strategies before action, while warning that visible chain-of-thought may be incomplete, fragile or deliberately concealed.
OpenAI turns Codex into an internal reliability engineer for ChatGPT
OpenAI is using Codex to investigate ChatGPT performance regressions, combining production telemetry, traces and testing workflows in an effort to accelerate reliability engineering and protect service speed.
Google’s Mac Agent Puts Gemini in the Driver’s Seat
Google is giving Gemini control of Mac apps, files and websites, testing whether AI agents can automate everyday work without sacrificing user oversight, security or reliability.
Apple’s Next AI Could Take Over the Mac
Apple is reportedly developing a computer-use AI agent for the Mac that could browse, control apps, run code and complete multi-step tasks, while raising major privacy, security and competition concerns.
AI’s Next Model War Is About Decisions, Not Conversations
AI’s emerging model war is moving beyond chatbots as Jev and OpenAI’s Decisions API target routing, moderation and other structured choices, raising new questions about cost, speed, calibration and trust.
Qwen4-Max Cuts AI Task Costs, but Gives Up Qwen’s Open-Model Edge
Alibaba’s Qwen4-Max targets enterprise AI with a reported 62% task-cost reduction, but its proprietary, hosted release sacrifices the open-model flexibility, deployment control and customization that made Qwen distinctive.