AI Daily: Gemini 4 Arrives Gated, FTC Pounces (Oct 2, 2026)
Google ships Gemini 4 Argon behind a trust gate, FTC probes OpenAI and Anthropic, OpenAI launches Dots, and Shopify lets you build stores by chat.
Two things landed at once: the FTC opened a probe into OpenAI, Anthropic and other AI companies over product risks, and Google revealed Gemini 4 Argon, its strongest model yet, behind an access gate. Meanwhile OpenAI used DevDay to launch the Dots agent on GPT-6 Astra. Read the terms you agreed to, then go ship.
1. FTC Opens a Probe Into OpenAI and Anthropic — Read Your Own Terms Before It Reads Them for You
The FTC is investigating OpenAI, Anthropic and other AI companies over product risks. If your product resells or wraps their APIs, your risk surface just moved too. Pull up your privacy policy, map where user data flows, and cut any capability claim you cannot back up. Builders who can show their work keep customers when regulators start asking questions.
Source: Hacker News
2. Gemini 4 Argon Is Google's Best Model Yet — and Most Builders Cannot Have It
Google revealed Gemini 4 Argon, pitched as frontier performance across real-world software engineering, enterprise knowledge work and cybersecurity. Access is limited to 'trusted cyber defenders' for now. Assume access gates like this are the new normal: keep your model layer swappable, test a second provider against your real workloads, and make the switch a config change instead of a rewrite.
Source: The Verge AI
3. OpenAI's DevDay Bet: Dots, an Agent on GPT-6 Astra, Aimed Straight at Meta's Muse
At its annual DevDay, OpenAI announced Dots, an agent powered by GPT-6 Astra, going head to head with Meta's Muse platform. Agent infrastructure is collapsing into a two-horse race with aggressive pricing on both sides. Keep your prompts, tools and memory in your own repo so you can repoint at whichever platform is cheapest next quarter.
Source: The Verge AI
4. Shopify Canvas Builds Stores by Chat — Clients Now Expect One in an Afternoon
Shopify launched Canvas, letting merchants build and customize a store by chatting with its Sidekick agent while changes render in real time. Client expectations just reset to same-day. Compete where a generic chat builder breaks: custom checkout logic, third-party integrations, and the messy edge cases that eat a weekend.
Source: TechCrunch AI
5. ChatGPT Does Virtual Try-On and Favorites Now — Discovery Is Moving Inside the Chat
OpenAI is rolling out shopping features in ChatGPT: virtual try-on using a user's own photos plus a Favorites library for saved products. If you sell anything physical, your catalog needs to be readable and structured for an assistant to recommend. The storefront is turning into a conversation, and the sellers who show up in it take the sale.
Source: TechCrunch AI
6. AWS Ships Strands Decider 2B — a Tiny Decision Model You Can Afford to Run
Amazon Web Services' Strand Labs released Strands Decider 2B, its latest Jevalike decision model, as decision models flood the web. A 2B model means cheap routing, classification and agent-policy calls on hardware you control. If you are burning frontier-model tokens on 'should I do X or Y' logic, swap that layer out this week and watch the bill drop.
Source: TechCrunch AI
7. ArXiv Tightened Its Rate Limits — Fix Your Paper Pipeline Before It Fails Silently
ArXiv published an updated rate limit policy. Anyone pulling papers for RAG, evals or fine-tuning datasets will hit it. Add caching, exponential backoff and a local mirror now, because a scheduled ingestion job that dies quietly at 3 a.m. is worse than one that never ran.
Source: Hacker News
Frontier access is gated, agent platforms are consolidating into two horses, and regulators are reading terms of service out loud. Pick the smallest thing on your roadmap, make your model layer portable, and ship it today.