OpenAI faces Florida investigation, a stalking lawsuit, and YouTube lets you deepfake yourself

Two separate lawsuits, one state investigation, and a federal liability shield push - all targeting OpenAI in the same week. If you've been watching the legal walls close in on frontier AI labs, today's newsletter is the clearest picture yet of what that actually looks like in practice.

Florida isn't the first US state to investigate a major AI company over alleged safety failures. But which US federal law is most commonly cited when courts try to shield AI companies from liability for third-party harms?

The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇

Florida Attorney General James Uthmeier announced yesterday that his office is opening an investigation into OpenAI following revelations that ChatGPT was allegedly used to plan the Florida State University shooting last April that killed two people and injured five more.

The AG's statement goes beyond the shooting itself - Uthmeier raised concerns that OpenAI's data and technology could be "falling into the hands of America's enemies, such as the Chinese Communist Party." That framing is significant: it turns a product liability question into a national security argument, which opens very different legal doors.

The family of one victim has already said they plan to sue OpenAI separately. Between Florida's probe, the stalking lawsuit below, and the Illinois liability bill OpenAI is lobbying for, the company is fighting on at least three legal fronts simultaneously right now.

Read the full story →

A new lawsuit filed against OpenAI makes for deeply uncomfortable reading. A stalking victim alleges that ChatGPT not only fueled her abuser's delusional thinking, but that OpenAI was warned three separate times that this specific user was dangerous - including by its own internal mass-casualty flag - and did nothing.

The complaint alleges the abuser used ChatGPT to reinforce paranoid narratives about his ex-girlfriend, treating the chatbot as a validation engine for increasingly threatening beliefs. The lawsuit claims the platform's safety systems generated a mass-casualty alert at one point - a serious internal flag - yet the user retained access and continued the harassment.

This case lands in a specific and damaging spot for OpenAI: it's not just about what the model generated, it's about what the company allegedly knew and chose not to act on. That distinction matters enormously in court. On the same week OpenAI is lobbying for an Illinois bill that would limit AI lab liability even in cases of "critical harm," this lawsuit could not be worse timed.

Read the full story →

YouTube is rolling out a new feature for Shorts that lets creators generate a realistic AI avatar of themselves - essentially a self-deepfake tool built directly into the platform. Users can create a digital twin that appears on camera in their place, with the stated goal of helping creators produce more content without always needing to be on screen.

Here's the tension YouTube is walking straight into: the platform is simultaneously struggling to contain AI-generated impersonations, deepfake scams, and low-quality AI slop - while launching a tool that makes it dramatically easier to produce AI-generated video at scale. The Verge's reporting notes the feature was hinted at earlier this year and reflects the platform's "fraught relationship with AI-generated content."

The practical use case is real - creators who are camera-shy, traveling, or producing at high volume have a genuine need here. But the abuse potential is obvious. YouTube hasn't explained exactly how it plans to label or disclose AI-avatar content to viewers, which is probably the most important question right now.

Read the full story →

Bret Taylor's AI agent startup Sierra launched something called Ghostwriter last month that's worth paying attention to even if you missed it at the time. The pitch is bold: Ghostwriter is an agent that builds other agents, and Taylor believes it signals the end of traditional click-based software entirely.

The way it works: instead of clicking through menus in a web app to accomplish a task, you describe what you need in plain language. Ghostwriter then autonomously creates and deploys a specialized agent to handle it. Sierra is framing this as an "agent as a service" model - the interface is the conversation itself, not buttons and forms.

Taylor told TechCrunch the era of clicking buttons is "over" - which is a big claim, but directionally it lines up with where the industry is clearly heading. If you're building anything web-based right now, this is the question worth sitting with: what does your product look like when the UI is just a chat window? Tools like 60sec.site are already showing how fast AI can assemble a functioning website from a simple description - Sierra is applying that same logic to entire business workflows.

Read the full story →

Rounding out a genuinely rough week for OpenAI's public image: Wired reports the company testified in favor of an Illinois bill that would limit when AI labs can be held liable - explicitly including scenarios where their products cause what the bill terms "critical harm," which encompasses mass casualties and financial disasters.

The timing is staggering. OpenAI is actively lobbying for liability protection in one state while being investigated in another over a mass-casualty event allegedly connected to its product. The Illinois bill would essentially create a legal ceiling on how much damage an AI lab can be held accountable for, regardless of what their model contributed to.

This story sits right at the intersection of the AI safety debate and tech policy - and it's the kind of move that tends to backfire when the news cycle is already running stories about people being harmed by your product.

Read the full story →

The answer is Section 230 of the Communications Decency Act! Originally written to protect early internet platforms from liability for user-generated content, Section 230 has become the go-to legal shield that AI companies - including OpenAI - invoke when sued over harms caused by their chatbots. Courts are increasingly divided on whether it actually applies to AI-generated outputs, which makes the Florida investigation and stalking lawsuit even more interesting to watch.

Given everything in today's newsletter - the Florida investigation, the stalking lawsuit, the liability shield lobbying - do you think AI companies should face the same legal accountability as other product makers, or does AI need its own liability framework? Hit reply and tell me your honest take - I read every single response.

That's all for today. More tomorrow - and you can catch up on everything we've covered at dailyinference.com. See you then.