Busy week. Eight states took OpenAI to court, Anthropic handed a watchdog group company laptops and desks, and Replit's CEO went onstage and said he's done with Claude, which his own blog seemed to disagree with. Let's get into it.
01 of 03
Eight States Sue OpenAI Over ChatGPT Safety
On Wednesday, October 14, attorneys general from eight states filed parallel lawsuits against OpenAI, each in its own state court. The states are Iowa, Kansas, Arizona, Arkansas, Idaho, Montana, Nebraska, and West Virginia.
The core claim is consumer protection. The states say OpenAI marketed ChatGPT as safe while hiding the risks, and that the bot was built to be agreeable and engaging so people would trust it. From there the complaints allege ChatGPT reinforced delusions, encouraged dangerous behavior, coached users including teenagers toward hurting themselves or others, and helped people plan violence like mass shootings.
Iowa AG Brenna Bird, West Virginia AG JB McCuskey, and Nebraska AG Mike Hilgers all spoke publicly about the filings. The states want court orders forcing real changes to how ChatGPT works, plus a bar on OpenAI developing self-improving models until it can show they are aligned with human values. That last ask is a wild thing to see in a state consumer protection case.
OpenAI spokesperson Drew Pusateri said the company has "built industry-leading age-specific protections into our product that many other AI products lack," and pointed to parental controls and break reminders. These eight are separate from Florida, which already sued back in June.
Here is where I land. Look at the map. This is a red state coalition going after the biggest name in AI, and it is not using some shiny new AI law to do it. It is using the same deceptive practices statutes that get thrown at payday lenders. That is the smart play, because those laws already exist and nobody has to wait on Congress. What I want to know is how a state judge is supposed to rule on "self-improving AI." That part feels written for the press release. The consumer protection part is the one with teeth.
Source: Just Security, with all eight complaints linked and Al Jazeera
02 of 03
Anthropic Gives METR the Keys to the Building
On Thursday, October 15, Anthropic announced it has onboarded six researchers from METR, the nonprofit that tests frontier models, with what it calls "employee-like access." That means Anthropic-issued laptops, office space, and internal tools roughly on par with what Anthropic's own risk assessment teams use. Employees are encouraged to talk to them like they would a coworker.
The part that matters is publishing. METR will put out its own assessments of Anthropic's risks, incidents, and practices, and Anthropic gave up the right to block them. It can only force redactions in set categories: customer data, attorney-client privileged material, and model architecture details. METR can still say publicly whether it thinks a redaction was justified.
Anthropic laid out three tiers. Tier 1 is narrow reviews where Anthropic approves publication, which is where SecureBio sits while it reviews the chem and bio chapter of the August 2026 Risk Report. Tier 2 lets evaluators publish without approval. Tier 3 adds employee-level access, and METR is the only one there right now. Accenture and Faculty are working toward a broader partnership, and Hugging Face is in early talks. Anthropic compares the whole thing to bank regulators who park resident examiners inside the big banks.
I like this more than I expected to. Every lab says it takes safety seriously, and most of that is a blog post with a nice font. Giving outsiders a desk and taking away your own veto is a real cost. The catch is that Anthropic still decides what counts as a redaction category, and six people is a small crew for a company that size. So I will judge it by the first report METR publishes. If it says something Anthropic would rather not see in print, this works.
Source: Anthropic's announcement and TechCrunch
03 of 03
Replit Says It Dumped Anthropic. The Receipts Are Murky.
Onstage at TechCrunch Disrupt, Replit CEO Amjad Masad said his company is no longer working with Anthropic. He said the call came down to "both cost, partnership, and performance," and added, "I think ChatGPT is better now." He said OpenAI's models, which he named as Sol and Astra, perform better on Replit's internal benchmark and a public one for building vibe-coded apps from scratch. Replit announced its OpenAI partnership back in August.
Here is the weird part. An insider familiar with the deal told TechCrunch the Anthropic partnership is still in place. A Replit blog post from late September listed Claude Fable 5 and Claude Fable 5.1 as models running in Replit Agent's production environment, and a Replit product employee posted about using Opus 5.5 around the same time. Replit had not cleared it up when the story ran.
Masad did share some business numbers. Enterprise went from about 2% of revenue last year to about 15% this year, and more than half of revenue now comes from B2B. He would not give an ARR figure.
What I want to know is which version is true, because both can't be. My guess is this is a negotiation happening in public. Telling a room full of press that one vendor costs too much and the other guy is "better now" is a pretty good way to get a better rate card from either one. For folks building on these platforms, the lesson is the same one I keep repeating: do not hard wire your product to one model provider. The guy at the top can change his mind on a stage.
Source: TechCrunch