BitsMinds

Microsoft Ships Computer-Using Agents in Copilot Studio to GA — AI That Drives Any App, No API Needed

Microsoft has made computer-using agents in Copilot Studio generally available across every commercial region, becoming the first major hyperscaler to put production-grade "computer use" into enterprises' hands. The agents operate desktop and web apps directly through the interface — clicking, typing and navigating like a person — using OpenAI's CUA and Anthropic's Claude Sonnet 4.5, wrapped in Azure Key Vault, Microsoft Purview auditing and human-in-the-loop review.

Microsoft

Altman and Amodei Walk Back Their AI “Jobs Apocalypse” — “I’m Delighted to Be Wrong,” Says Altman, as Both Eye $1T IPOs

On May 26, both Sam Altman and Dario Amodei softened their AI “jobs apocalypse” warnings. Altman told Commonwealth Bank’s CEO he was “pretty wrong” and “delighted to be wrong” that entry-level white-collar roles haven’t been wiped out, while Amodei recast his 50%-of-white-collar-jobs warning as automation that “expands” human work. The walk-back lands just as OpenAI and Anthropic head toward IPOs valued near $1 trillion each.

Fortune / Time

Anthropic’s Project Glasswing Turns an Unreleased Claude Mythos Loose on Open Source — and Finds 10,000+ Zero-Days in a Month

In a May 26 progress update, Anthropic said Project Glasswing has used an unreleased frontier model, Claude Mythos Preview, to autonomously discover more than 10,000 high- and critical-severity zero-day vulnerabilities across the world’s most important software in its first month — including a 27-year-old flaw in OpenBSD and a certificate-forgery bug in wolfSSL. The model is so capable at offensive security that Anthropic is refusing to release it.

Anthropic / Help Net Security

Beijing Extends AI Travel Curbs to Private Firms, Putting Alibaba, DeepSeek and Others on the No-Fly List Without Approval

Bloomberg reported on May 26 that China has quietly extended the overseas-travel restrictions long applied to state researchers and nuclear scientists to engineers, founders and executives at private AI companies including Alibaba and DeepSeek. Selection is by "strategic value" — not seniority — and Beijing is openly framing the move as a counter to Meta's $100M signing bonuses and DeepSeek's Singapore relocation.

Bloomberg / Tom's Hardware

Paul Graham Says AI-Written Founder Emails "Feel Like Being Lied To" — and He Stops Reading the Moment He Spots One

In a viral X post on May 25, Y Combinator co-founder Paul Graham drew a line under the cold-outreach gold rush: he can spot AI-written founder emails by their "hard-hitting journalistic style," he stops reading the moment he notices, and the practice "feels like being lied to." A million views and a wave of reactions from Mike Isaac, Kenneth Stanley and Steven Sinofsky later, hand-written email is suddenly the new premium signal in a VC inbox.

Digg / BeInCrypto

Inside GenAI.mil: 25 Pentagon "Power Users" Are Bake-Offing OpenAI, Google and Grok Against Claude

Bloomberg pulled back the curtain on the DoD’s replacement search for Claude this week: a hand-picked group of 25 military "power users," a new platform called GenAI.mil that sits independent of Maven Smart System, and a six-month transition deadline for every contractor that built on Anthropic. Frontier models from OpenAI, Google and xAI are now being graded inside classified workflows that Claude used to own.

Bloomberg / The Star

Together AI Open-Sources OSCAR, a 2-Bit KV Cache That Cuts Long-Context Serving Costs in Half

OSCAR — Offline Spectral Covariance-Aware Rotation — squeezes the KV cache that bloats every long-context request down to two bits per element, with no measurable accuracy loss on Qwen3-32B or GLM-4.7 across reasoning, math and coding benchmarks. Together AI shipped it into SGLang on May 25, framing it as one of the more practical wins for actually-affordable 1M-token serving.

MarkTechPost / Together AI

TeamPCP Hackers Put 450 Mistral AI Repos and 5GB of Source Code Up for Sale at $25,000

The same supply chain crew that hit TanStack and rotated OpenAI’s code signing certificates is now auctioning what it says is nearly 450 internal Mistral AI repositories — training, fine-tuning, benchmarking, model delivery and inference code — with a one-week deadline before a free dump on hacker forums. Mistral confirms a breach, says its hosted services and user data were not touched.

BleepingComputer / TechRadar

Anthropic's Leaked Forbidden-Strings File Names Opus 4.8, Sonnet 4.8 and a New Tier Above Opus Called Capybara

Inside the 512K-line Claude Code source map that npm shipped on March 31 was a feature literally called Undercover Mode — and inside Undercover Mode was a hard-coded list of names it was supposed to scrub from public builds. That list spells out Opus 4.8, Sonnet 4.8, a never-released Sonnet 4.7 and animal codenames Capybara and Tengu — with Capybara described internally as a new tier sitting above Opus.

WaveSpeed AI / ChaoBro / AI Nexus Daily

ComfyUI: The Node-Based Workflow That Power Users Actually Run

Runway Gen-4.5: The Production Standard for AI Video in 2026

Lovable: Building Full Apps Without Writing Code in 2026

v0 by Vercel: Prompt-to-React UI With Real Components You Can Ship