🔓 The AI too dangerous to release got out anyway

🖥️ Microsoft rolls out Agent Mode across Word, Excel, and PowerPoint

🤖 OpenAI launches workspace agents in ChatGPT

💰 Elon Musk's $29B consolation prize just got taken back

🤖 UK construction site hires a robot for admin work

+ 📊 Daily poll and results

+ 📈 Trending tools and resources

1,000+ ready-to-use prompts to solve problems in minutes instead of hours—tested & used by 1M+ professionals

Superhuman AI newsletter (3 min daily) so you keep learning new AI tools & tutorials to stay ahead in your career—the prompts are just the beginning

A Discord group accessed Anthropic's Mythos model on April 7th, the same day it launched privately, using a contractor's credentials and Mercor breach data to guess its location online.

Mythos can identify and exploit vulnerabilities across every major OS and browser. Anthropic restricted it to Project Glasswing partners like Nvidia, Apple, and Microsoft over weaponization concerns.

The group has been using Mythos regularly and even gave Bloomberg a live demo. Anthropic says it found no evidence of system-level impact, but is still investigating the third-party vendor breach.

Microsoft launched Agent Mode in its Office apps this week, letting Copilot directly edit documents, build Excel formulas, and update PowerPoint decks in real time instead of just answering questions.

A Microsoft VP admitted the original Copilot couldn't act on documents because foundation models weren't powerful enough then. Better reasoning and instruction-following in newer models made Agent Mode possible now.

Microsoft is calling this "vibe working," which is either inspired branding or a sign that someone in Redmond has been on LinkedIn too long. The feature rolls out to all Microsoft 365 tiers.

OpenAI introduced workspace agents in ChatGPT, letting teams build and share AI agents that handle complex workflows like drafting reports, writing code, and responding to messages across an organization.

The agents run in the cloud, connect to external tools like Slack, and can request human approval mid-task. They are powered by Codex and available now in Business, Enterprise, Edu, and Teachers plans.

OpenAI's own sales team already uses one to qualify leads and draft follow-up emails. GPTs never quite took off as a product, and this looks like OpenAI's second attempt to make them stick.

Tesla's board voted on April 21 to revoke the $29 billion interim package it gave Musk in August 2025, after the Delaware Supreme Court restored his original 2018 award worth $56 billion.

Tesla had designed the interim package as a hedge against losing the appeal, with a written "no double dip" commitment to shareholders that it would be cancelled if Musk prevailed in court.

The revocation leaves Musk's $1 trillion package untouched, though Tesla's own filing suggests it thinks several of the operational milestones tied to that award are unlikely to be hit.

Tilbury Douglas became the first major UK builder to use a humanoid robot on a live site. Named Douglas, the Unitree-built bot handles 360-degree photo capture, progress tracking, and safety monitoring autonomously.

Douglas saves roughly 40 hours of admin work per month by hitting the exact same coordinates daily, letting AI software flag deviations and defects that human walkers might miss.

At £15,000, Douglas costs less than many annual software licenses. The company says it's a tool to free up workers, not replace them, which is exactly what companies say right before they order more robots.

A) No, that's a dealbreaker - 94% 🏆

B) Yes, if it's a big company like Meta - 6%