Eight states just sued OpenAI over child safety
Issue #059 · 4 min read
Eight states just sued OpenAI over child safety
Plus Figma defines how agents work on its platform, and Gemini 4 Argon lands everywhere at once
The big story
Eight states have sued OpenAI, arguing ChatGPT isn't safe for kids — that's The Verge.
This isn't a one-off.
Platformer lays out the bigger pattern: state attorneys general already beat Meta in New Mexico using a similar playbook, and now they're pointing the same legal machinery at OpenAI.
That matters because it's not one state with one grievance.
It's eight, at once, using a strategy that already won once.
If it works again, the theory of liability doesn't stay contained to OpenAI. It becomes the template every state AG reaches for against any company shipping a chatbot to minors.
For product teams, the practical question isn't "is our model safe." It's "can we show our work." Expect procurement and legal to start asking for documented safety testing and age-gating decisions the way they already ask for SOC 2 reports — not because they care about the paperwork, but because paperwork is what holds up in court.
What shipped
Figma is defining how agents work on its platform, and shipping code layers in open beta — TechCrunch.
If you've been waiting for Figma to say where agents are and aren't allowed to touch a design file, this is that answer starting to take shape. Worth a look if your design team already has agents poking around files unsupervised.
Gemini 4 Argon is now everywhere a developer might look for it. It's generally available in GitHub Copilot, live on GitLab Duo Agent Platform for long-running agentic work, and it's Google's new frontier model launch in its own right.
Three platforms shipping the same model in the same week tells you where the integration race actually is right now: not in training better models, but in who gets a frontier model into developers' hands first.
GitHub shipped a run of smaller but real changes for anyone managing agent access at scale: workflow lockfiles stop a moved tag or branch from silently changing what your CI runs, GitHub App permissions got tighter defaults to help you off classic personal access tokens, and secret scanning now blocks Anthropic keys by default and adds Salesforce detection.
None of these are exciting on their own. Together they're GitHub quietly building the guardrails for a world where agents, not people, are the ones holding credentials.
What I'd actually do this week
Pull up your own product's safety testing documentation and ask whether it would survive a state AG's discovery request — not a security audit, a legal one. The gap between those two is where the OpenAI suits are aimed.
If you're running agents against GitHub or GitLab, check whether your team is still using classic PATs. The new least-privilege App permissions exist because that's the thing attackers go after first.
Read the Figma announcement if any part of your design-to-code handoff already involves an agent you didn't explicitly configure. You may find it's been making decisions nobody signed off on.
Hit reply if you're seeing the state AG lawsuits show up in your own compliance conversations — I'd like to hear what procurement is actually asking for.