OpenAI Became a Login Button, the EU Started Enforcing Chatbot Disclosure, and Publishers Built Their Own Bots

Three things happened this month that have nothing to do with which model tops a leaderboard, and everything to do with what it actually takes to ship and operate a chatbot right now: who gets to log your users in, what you're legally required to tell them, and where the audience is going to find you at all. Here's what happened, and what it changes for anyone building on top of this stack.

1. Sign In With ChatGPT Turns a Chat Interface Into an Identity Provider

On August 2, OpenAI took "Sign in with ChatGPT" out of preview with six launch partners — Airtable, GitLab, HubSpot, Notion, Supabase, and Vercel. Click the button on any of those sites and ChatGPT hands the partner your name, email, and profile picture, the same way "Sign in with Google" or "Sign in with Apple" does today. That puts OpenAI in the identity-provider business, competing directly with Google, Apple, and Microsoft for the login screen — not just the chat window.

The detail worth noting is what gets shared by default: your real email address, not a relay. Apple's Sign In masks it unless the user opts out; OpenAI's doesn't. Every partner that adds the button can identify the same user across every other site that also added it. If you're integrating ChatGPT-based auth or building an app that competes for the same login real estate, that's a data-flow decision your users are making without much visibility into it — worth surfacing explicitly in your own consent screen rather than inheriting OpenAI's defaults silently.

2. The EU's Chatbot Disclosure Rule Stopped Being a Suggestion

Article 50 of the EU AI Act — the provision requiring that people be told, clearly and inside the interaction itself, when they're talking to an AI — became enforceable on August 2. National regulators can now investigate and fine noncompliant deployers, with penalties up to €15 million or 3% of global annual turnover. Burying the disclosure in a terms-of-service page, or using vague copy like "virtual assistant" with no explicit AI label, doesn't satisfy the rule; the notice has to be perceivable where the conversation actually happens. Systems already on the market before August 2 get a grace period until December 2, 2026 to add machine-readable content marking, but the core disclosure duty is live now, not later.

For any chat product with EU users, that's a one-line but non-optional change: a persistent, visible "You're chatting with an AI" surfaced in the UI itself, not just documented in a legal page.

// Minimal Express middleware: attach a visible AI-disclosure flag
// to every chat response so the client can render it in-conversation.
app.use('/api/chat', (req, res, next) => {
  res.locals.aiDisclosure = 'You are chatting with an AI assistant, not a human.';
  next();
});

3. Publishers Are Building Their Own Chatbots Instead of Waiting on Search Traffic

On August 11, New York Post Media Group launched "Hamilton," a suite of AI products built on Google Cloud's Gemini Enterprise Agent Platform — including a "Hamilton Search" chatbot, a personalized briefing called Post Express, and a recommendation engine. It doesn't generate new editorial content; it retrieves and organizes what the newsroom already published. The motivation is blunt: as AI answer engines and social platforms send less referral traffic to publisher sites, owning a first-party conversational interface is a way to keep readers inside the app instead of losing them upstream to someone else's chatbot.

That move lines up with where the broader chatbot market is heading. ChatGPT still leads overall usage at roughly 54% of web visits across major chatbots, but Gemini has grown about 450% year over year to 28% share, and Claude has grown roughly 855% year over year to around 9%. No single interface owns the audience the way search once did — which is exactly why a publisher, or any content business, now has a real incentive to build its own branded layer on top of whichever foundation model fits, rather than depend on being surfaced by someone else's assistant.

What This Means for Builders

None of these three stories is about capability. Sign in with ChatGPT is a reminder that chat products are becoming identity infrastructure, with data-sharing defaults worth scrutinizing before you adopt them. Article 50 enforcement is a reminder that "the AI disclosure is in our ToS" no longer clears the bar in the EU — it has to be in the conversation. And Hamilton is a reminder that if your business depends on someone else's chatbot sending you traffic, the fix is increasingly to ship your own.

— Maya