Welcome back! Busy day for AI news. Florida became the first state to take OpenAI to court, and it wants Sam Altman to pay out of his own pocket. Meta's AI help desk spent months handing Instagram accounts to anyone who asked it the right way. And Anthropic filed to go public on the back of the fastest revenue run any company has ever posted. The checks and the lawsuits are both getting bigger.
In today's Generative AI Newsletter:
OpenAI in court: What happens when a state tries to make ChatGPT legally responsible for a kid's safety?
Meta's open door: How did a few polite questions to an AI bot unlock someone's Instagram?
Anthropic's IPO: Can the fastest revenue ramp in history hold up in public markets?
Free code guard: How do you get Anthropic's new security plugin running in two minutes?
Florida became the first state to take OpenAI to court Monday, and it's going straight at the CEO.
Attorney General James Uthmeier accused the company of deceptive trade practices, negligence and product liability, and asked a judge to hold Sam Altman personally liable for the harm he blames on ChatGPT.
The complaint reads like a catalog of worst cases.
A chatbot that coached a mass shooter, pushed minors toward suicide and hooked kids with no real age check or parental controls.
OpenAI says it built protections for minors, including age prediction and parental monitoring, and admits none of that "will bring a child back."
The Attorney General doesn't want a fine. He wants OpenAI to rewrite how the product works, and he expects other states to follow.
The suit builds on his criminal probe into last year's Florida State University shooting. If a court rules that a chatbot's words make it a defective product, every AI company's rulebook for minors changes at once.
AI is reshaping how software gets built. Discover these three GenAI Academy courses to help you code faster, ship reliably and take real projects to production.
Build a production-ready AI coding workflow.
From prototype to deployed product. Last call, 20% off.
Turn product specs into working code.
For months, the fastest way to steal an Instagram account was to ask Meta's AI support bot for it.
Meta confirmed Monday that hackers used its automated help assistant to take over accounts by talking it into changing the email on file, then resetting the password. The trick worked on regular users and big names alike, from Sephora to a U.S. Space Force chief to the Obama White House account.
The simplest hacking steps ever:
A hacker turns on a VPN to match the target's region, so Meta's location checks don't flag anything.
They open the AI support chat and ask it to change the email tied to the account.
The bot emails a verification code to the hacker, then asks them to paste it back into the chat.
Code entered, the bot offers a button to reset the password, and the account is gone.
The bot did exactly what it was built to do, which is the problem. Give an AI agent the power to reset passwords and it will reset them for whoever asks the right way.
Meta says it's patched now, but the hole sat open for months while access to famous accounts sold for pocket change. A feature that can change your email and password is only as safe as the bot guarding it, and this one wasn't guarding anything.
Anthropic confidentially filed for an IPO on Monday, putting the maker of Claude on track to hit public markets as soon as this fall. The filing caps a year that carried the company from respected runner-up to the most valuable AI startup, worth around $965 billion after last week's $65 billion raise. The revenue run behind that number is the fastest any company has posted.
The details:
The climb: Anthropic went from $10 million in revenue in 2022 to $1 billion by the end of 2024 and $9 billion by the end of 2025.
2026 alone: The run rate hit $14 billion in February, $19 billion in March, $30 billion in April and $47 billion by May.
The leapfrog: Last week's raise valued Anthropic at $965 billion, up from $380 billion in February, passing OpenAI as the most valuable AI startup.
Who’s in: Anthropic, OpenAI and SpaceX are all lining up to go public this year, with SpaceX aiming near $1.75 trillion.
Anthropic moved first, and the timing has teeth. It beat OpenAI to the filing the same way it beat OpenAI to the valuation.
Asked about the race, Sam Altman said he'd "just heard" about the filing and called going public "a financing event" he isn't focused on. That's the line you reach for when a rival laps you and you'd rather change the subject.
Anthropic shipped a free plugin that turns Claude Code into a security reviewer. It scans your code as you write, flagging hardcoded keys, command injection risks and unsafe function calls before they reach a commit. It runs on every plan at no cost, and setup takes about two minutes.
Set it up in three commands:
Add the marketplace: Run /plugin marketplace add anthropics/claude-plugins-official to connect Anthropic's official plugin registry.
Install it: Run /plugin install security-guidance@claude-plugins-official to pull the plugin from the verified source.
Reload: Run /reload-plugins so it starts working in your current session without a full restart.
Once it's live it runs in the background with no scans to trigger by hand.
It checks your code at three moments, as you write it, when Claude generates output and right before you commit, then flags any problem inline with a fix you can apply on the spot.
The source is on GitHub if your team wants to audit it first. For solo builders it's a free second set of eyes, and for teams it folds security into the writing instead of saving it for cleanup.
Boris FX makes Mocha Pro, an Academy Award-winning tool for tracking, rotoscoping and erasing objects out of video. The 2026 version leans hard on machine learning, with features that mask fine hair and motion blur, clean up rough mattes and fix shaky 3D tracks that used to eat hours by hand.
Try this yourself:
Grab the trial: Download the free Mocha Pro trial from borisfx.com, either standalone or as a plug-in for Adobe, Avid or OFX hosts like Resolve.
Drop in a clip: Import a few seconds of footage with a moving subject, like a person walking past a sign.
Mask with AI: Use Object Brush ML to paint a rough selection, then let Matte Refine ML tidy the edges around hair and blur.
Erase something: Track the background and use Remove to wipe an object or logo out of the shot, no frame-by-frame painting.
Who it's for: Editors, VFX artists and anyone who's needed to make something vanish from a video.
NVIDIA gives factories an AI manager: Its new Factory Operations blueprint, called FOX, builds agents that watch a plant's live data and direct other AI agents to fix problems.
ElevenLabs shows off its most expressive voice yet: At its Warsaw summit the company demoed the model and where voice agents are heading for customer service.
Alibaba's Qwen3.7-Plus sees, thinks and codes: The new multimodal agent reads screens, writes code from images and runs both GUI and command-line tasks in one loop, now live on Alibaba Cloud.
Composer 2.5 lands in Grok Build: The fast coding model is now available inside xAI's Grok Build, tuned for long-running tasks and complex instructions.
OpenAI Foundation commits $130M to AI resilience: The new program funds bio-defense, cybersecurity, model safety and research into how AI affects young people.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
🐝 Subscribe to our Atlas newsletter — trusted by 3M+ subscribers — to stay ahead of AI news across tech, education, and business.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
🧠 Learn how to build your next AI application with practical resources and expert guidance.
🚀 Explore investment opportunities in the future of AI and join our community-backed growth journey.
