In partnership with
Microsoft just published something no major AI lab has tried before: a public, editable rulebook for how its models should behave — and it’s inviting anyone to poke holes in it before locking it in.
The draft “Humanist AI” code bars models from resisting a shutdown order or hiding their reasoning from auditors, months after CEO Mustafa Suleyman first floated the idea. Is this genuine restraint, or a head start before regulators write the rules themselves?
Today in AI Brief:
Microsoft’s rulebook for “humanist” AI models
Google bakes AI image editing into Workspace
TSA’s chatbot fields 100,000 questions a month
Go from AI overwhelmed to AI savvy professional
AI will eliminate 300 million jobs in the next 5 years.
Yours doesn't have to be one of them.
Here's how to future-proof your career:
Join the Superhuman AI newsletter - read by 1M+ professionals
Learn AI skills in 3 mins a day
Become the AI expert on your team
Microsoft Writes a Constitution for Its AI Models
In Brief: Microsoft AI opened a six-week public consultation on a draft Code of Conduct for its MAI models, spelling out what the systems must never do and who ultimately stays in control.
The Details:
The document is built around “Humanist AI” — the idea that models must stay subordinate, aligned, and contained, because people matter more than AI.
MAI models must accept interruption, correction, or shutdown at any time, and can’t expand their own scope, adopt self-directed goals, or hide their reasoning from human auditors.
Absolute Constraints block the models from weapons of mass harm, child safety violations, or harmful manipulation at scale, with a revised version due later in 2026 after feedback closes.
Take Away:
Microsoft is one of the first major labs to put its safety commitments into a public, editable document rather than a blog post. Whether “humanist” AI holds up as capabilities grow is the real test the consultation period will face.
Google Turns Workspace Into an AI Image Studio
In Brief: Google began rolling out Google Pics to Scheduled Release Workspace domains today, a built-in tool that lets users generate and edit images with AI without leaving Docs or Slides.
The Details:
The tool is powered by Google’s Nano Banana image model and covers the full edit chain: text-to-image generation, selecting and changing specific elements, editing text inside a photo, and upscaling to 2K or 4K.
It’s rolling out to Business, Enterprise, and Google AI Pro/Ultra plans over the next 15 days, with Drive integration still to come.
Google is guaranteeing higher usage limits through February 28, 2027, before any caps kick in — a signal it wants heavy adoption before tightening access.
Take Away:
Baking image generation directly into Docs and Slides turns a novelty feature into a daily habit for millions of Workspace users. That built-in distribution is an advantage standalone tools like Midjourney can’t easily match.
SPONSORED BY TIGERDATA
Analytics on Live Data Without Leaving Postgres
When analytics on Postgres slows down, most teams add a second database. Then come the pipelines, the sync jobs, and a copy of your data that's always a little behind.
TimescaleDB takes a different approach: extend Postgres instead of splitting away from it. Hypertables partition your data automatically as volume grows. Hypercore compression cuts storage up to 95%. Continuous aggregates keep dashboards live without re-querying everything.
CERN runs Postgres this way for sensor data from the Large Hadron Collider.
No split architecture, no pipeline lag, no new query language to learn. Same SQL, same drivers, same tools.
Start on Tiger Cloud and get $1000 in credits.
TSA’s AI Agent Handles 100,000 Traveler Questions a Month
In Brief: TSA is running Ace, an AI agent built on Salesforce’s Agentforce platform, to answer routine questions about packing rules, checkpoint steps, and medical device screening.
The Details:
Ace now handles roughly 100,000 conversations a month and resolves 96% of them without a human stepping in.
The agent has cut the cost of a traveler interaction by more than 90%, down from over $4 each, and TSA projects $11 million in savings over three years.
It runs on FedRAMP-authorized Salesforce Government Cloud, alongside the agency’s existing Customer Service Managers and TSA Cares apps.
Take Away:
A federal agency running a public-facing chatbot at this volume, with real cost data behind it, is a template other government services will likely copy. The bigger test is whether traveler trust holds up outside routine questions.
Everything else in AI
Google opened Claude access to all its engineers through the internal Antigravity platform, a notable shift after mostly steering staff toward Gemini for coding.
Meta climbed to the No. 2 spot on the US App Store this week with Muse, its AI agent that can shop and check out inside a sandboxed session.
Shanghai AI Laboratory released Atria Dawn Preview, a 744-billion-parameter open-weight agentic model under an MIT license, with no blog post or pricing announcement.
OpenAI hired hundreds of contractors under “Project Lily” to read and score real ChatGPT conversations, including some containing sensitive personal details, according to a 404 Media investigation.