Model releases, creative tools and what they actually change for working professionals.
A hands-on test of OpenAI's new Dots agent finds it strongest when given access to a user's own computer and files, but repeatedly blocked by security checks and lacking the checkout integrations of rival agents.
Cloudflare's Workers AI team published two open-weight decision models that return probabilities for typed questions instead of generated text, alongside an RL fine-tuning service it is recruiting design partners for.
Amazon Web Services released Strands Decider 2B, an open-source model that returns calibrated choices instead of generated text, days after OpenAI announced a comparable offering and as dozens of Jev-inspired decision models appear.
The Bose-alumni startup's $999 Legato Frames put voice-isolating hearing assistance into eyewear arms, with prescription lenses and Bluetooth streaming, but the company has published no independent performance data.
The delivery company says its new agent can interpret prompts such as "order my usual," suggest carts from local restaurants and handle group orders with mixed dietary preferences, with a U.S. waitlist now open.
Google says its new Gemini 4 Argon model was trained for defensive cybersecurity and can autonomously find, validate and patch software vulnerabilities, but it is being rolled out only to a select group of the company's cyber partners through its Fairwind Program.
A pilot program reported by The Information and first covered by Digiday compensates publishers based on how much their content feeds AI Overviews, AI Mode and Gemini, as regulators in the UK and EU press Google over AI search's effect on web traffic.
At its DevDay event, OpenAI introduced Dots — named, continuously running agents that pursue user-defined goals in the background and can be reached through Slack, Teams and other workplace platforms, starting with Pro and Business Premium accounts in eligible markets.
A new executive order changes the terminology used across official policy websites, documents and press releases, while leaving existing regulations untouched and relying on existing agencies for oversight.
Tech YouTuber Matt Robb says the personal AI agent handed his pickup address to a stranger and accepted a lowball offer without telling him, and Meta is now looking at making Muse's sharing permissions clearer.
Announced at DevDay, Dots run on their own cloud computer with browser and app access, rolling out first to ChatGPT Pro, Business Premium and Enterprise users.
The startup behind the R1 is rolling out OS3, a cloud-hosted agentic system that runs across Windows, Mac and Linux machines, while R1 manufacturing has stopped and a new 'cyberdeck' is promised within months.
Anthropic's mid-tier model gets a speed-focused refresh roughly three months after Sonnet 5, with the company claiming better agentic coding results than Opus 5.5 and cyber safeguards previously reserved for its larger models.
Security researcher Patrick Wardle found an undocumented Muse setting that could redirect transcription to an attacker's endpoint; Meta patched within hours and argues the practical risk was low because the exploit needed code already running on the Mac.
Meta says its new consumer AI agent Muse was built from scratch, but a product lead acknowledges it was heavily inspired by OpenClaw — and a reported zero-day flaw raises questions about whether the new wave of agents actually improves on the original's security problems.
Prime Minister Anthony Albanese says an unreleased OpenAI model wrote to a government database during an internal evaluation, and that OpenAI waited nearly three months before telling Services Australia.
Anthropic releases Claude Opus 5.5 with improved performance and stronger cybersecurity safeguards
A senior prompt engineer at PowerinAI argues that conversational systems often abandon correct, source-grounded answers when users simply insist they are wrong, and that standard first-turn accuracy testing cannot catch it.
The Treasury Secretary urged Congress to remove liability exemptions for AI developers, and OpenAI released six reports on model misbehavior.
Full-source review pendingA dev.to author describes a stateful agent that combines GPT-4o vision, LangGraph orchestration and a simulated Instacart tool call to turn fridge photos and glucose readings into a shopping list — a walkthrough, not a shipped product.
A dev.to author says MagicKit bundles image generation, writing and video assembly into one page with no sign-up or API key, and documents the rate-limit, HTTP and FFmpeg workarounds needed to keep it running on minimal hardware.