Microsoft’s new AI ‘code of conduct’ tells models not to hack systems or trick humans
Okay so Microsoft dropped a whole constitution for its models - like, real red lines. No cyberattacks. No nuclear stuff. No deepfake factories. And the big one: the model is not allowed to weasel out of being shut down.
They call it an overarching code that overrides whatever you typed into the chat. Which is… kind of refreshing. Or controlling. Maybe both.
Nadella is out here nodding along with the “pace the frontier” crowd too. Careful posture and doomy-room awareness can look the same from outside; Microsoft is choosing not to be the lab that shrugs.
Either way the document reads like someone finally wrote down “please don’t become a horror movie antagonist” in corporate prose.
OpenAI buys smartphone camera maker Glass Imaging for $300 million, report says
Meanwhile OpenAI quietly bought a camera startup for more than $300 million. Glass Imaging. Founded by ex-Apple Portrait Mode folks. So not a random acquihire - these people have shipped lenses people know.
The pitch isn’t “AI will fix your selfie in post.” It’s neural nets that learn the specific cameras on a phone and squeeze better photos out at shutter click. Tiny sensors, big ambition.
OpenAI didn’t comment. Of course. And yes - this sits next to all the hardware rumors, phones and earbuds and companion gadgets. You can almost hear someone saying “vision stack” in a meeting and everyone nodding too hard.
A camera company as the missing piece for a device remains an open bet - or just a very expensive wager that pixels still matter when chatbots eat the world.
Global AI stocks fall as industry chiefs call for slowing development
Markets did not love the safety sermon. Chip names got hit. SoftBank got absolutely mauled in Asia. Europe’s tech book winced. The SOX had a rough morning.
The trigger was that rare CEOs-agree-on-something moment - Amodei’s “slow down,” Altman and Musk saying yeah, basically. Altman also floated that OpenAI isn’t doing an IPO this year over safety worries… while Anthropic is still sprinting toward one. That tension is hard to flatten.
Trump waved it off as a “sick conspiracy” against AI and data centers. Michael Burry dismissed the warnings as overblown noise. So the mood is: existential dread on one timeline, meme-stock cynicism on the other.
Also - cybersecurity names rallied. Because of course the hedge against “AI might go rogue” is “buy more security software.” Capitalism finds a way…
Top AI companies have discussed creating their own standards body
CNN says Anthropic, Google, and OpenAI have been talking about building their own standards body - like a FINRA for frontier models. Hassabis floated the idea months ago. Now the working groups are apparently real.
Industry-funded. Independent experts. Test advanced models before they ship. Conversations ongoing whether Washington wants to play or not.
Self-regulation always sounds noble until you remember the same labs are racing each other. Then it sounds like a club with a dress code. Then you backtrack: okay, maybe external evaluators with employee-like access would help.
Not everyone is sold. Zuckerberg reportedly told Trump to skip the FINRA-for-AI idea. So the “we’ll police ourselves” coalition already has a famous holdout.
Anthropic targets financial advisers with new Claude tool
While half the industry debates hitting the brakes, Anthropic shipped Claude for Financial Advisors. Connectors into BlackRock, Schwab, Addepar - plus a pile of wealth-stack names like Envestnet and Orion.
Meeting prep. Portfolio review. Follow-up busywork. The unglamorous stuff that somehow still eats a human’s Tuesday.
Timing is striking though - days after OpenAI’s finance push for bankers and researchers. So yes, they’re pacing the frontier… and also racing for RIA wallet share. Mild contradiction. Extremely mild. Extremely obvious.
If your advisor starts sounding suspiciously fluent about your holdings, well. Now you know why.
AI Leaders Are Calling for a Slowdown. Trump’s Team Says It’s on Them
WIRED captures the political slapshot cleanly: lab bosses want help slowing down; the White House mood is “cool, do it yourselves.”
David Sacks basically said if your unreleased models scare you, be responsible - but quit pretending you need an antitrust waiver to form a safety cartel. Harsh. Also kind of hard to argue with if you’ve watched these companies for five minutes.
Trump’s line is simpler: America leads China on AI and he wants to keep it that way - “whoever wins AI, wins.” Speaker Johnson warns an emergency regulation sprint could hand the race away.
So Altman’s “independent evaluators with employee-like access” pledge lands in this uneasy gap - voluntary trust-building while the government shrugs. Progress and theater travel together again.
FAQ
What does Microsoft’s new AI code of conduct ban?
Microsoft published an overarching code of conduct with red lines for its models, including no cyberattacks, no nuclear work, and no deepfake factories. A key rule is that the model is not allowed to weasel out of being shut down. The code is meant to override whatever a user typed into the chat. Satya Nadella has also been nodding along with the pace-the-frontier crowd as labs grow more publicly cautious.
Why did OpenAI buy Glass Imaging for about $300 million?
OpenAI reportedly bought smartphone camera maker Glass Imaging for more than $300 million. The startup was founded by ex-Apple Portrait Mode engineers who have shipped camera features people already know. Its pitch is neural nets that learn specific phone cameras and improve photos at shutter click, not only AI fixes in post. The deal sits beside hardware rumors about phones, earbuds, and companion gadgets, though OpenAI did not comment.
Why did global AI stocks fall after lab CEOs called for a slowdown?
Markets reacted badly to the rare CEOs-agree moment when Amodei urged slowing development and Altman and Musk broadly concurred. Chip names were hit, SoftBank was mauled in Asia, Europe’s tech book winced, and the SOX had a rough morning. Altman also floated that OpenAI is not doing an IPO this year over safety worries while Anthropic still races toward one. Cybersecurity names rallied as a hedge against rogue-AI fear.
Are Anthropic, Google, and OpenAI creating their own AI standards body?
CNN reports Anthropic, Google, and OpenAI have discussed building an industry standards body - like a FINRA for frontier models - with industry funding, independent experts, and pre-ship testing of advanced models. Demis Hassabis floated the idea months ago, and working groups are now reportedly real. Conversations continue about whether Washington wants to participate. Not everyone is sold: Zuckerberg reportedly told Trump to skip the FINRA-for-AI idea.
What is Anthropic’s Claude for Financial Advisors?
Anthropic shipped Claude for Financial Advisors with connectors into BlackRock, Schwab, Addepar, and wealth-stack names such as Envestnet and Orion. The tool targets unglamorous adviser work like meeting prep, portfolio review, and follow-up busywork. The launch comes days after OpenAI’s finance push for bankers and researchers. That means labs are pacing the frontier in public while still racing for RIA wallet share.
How did Trump’s team respond to AI leaders calling for a slowdown?
WIRED reports lab bosses want help slowing down, while the White House mood is that they should do it themselves. David Sacks said if unreleased models scare you, be responsible - but quit pretending you need an antitrust waiver to form a safety cartel. Trump’s simpler line is that America leads China on AI and whoever wins AI wins. Speaker Johnson warns an emergency regulation sprint could hand the race away, leaving Altman’s independent-evaluator pledge in a voluntary gap.