Friday’s through-line is practical AI you can touch today, a coding agent you can reshape, and a safety fight inside OpenAI that spilled into headlines. ChatGPT launched a global virtual try-on and Favorites library for shopping. Anthropic shipped Claude Code mods so teams can customize the agent in TypeScript. And OpenAI parted ways with three researchers over alleged mishandling of sensitive information, while California’s attorney general sent the company an investigative subpoena.
1. ChatGPT can virtually try clothes on you (and save what you like)
OpenAI rolled out two shopping upgrades globally on Thursday: Virtual Try On and Favorites. Per TechCrunch and Engadget, when clothing shows up in ChatGPT shopping results you can tap Try On, upload a selfie or full-body photo, and get a generated preview of yourself in the item. You can also upload a screenshot of a garment and ask ChatGPT to try it on. Favorites lets you park products (and try-on images) in a Library inside the app for later.
OpenAI says the feature uses ChatGPT Images 2.5, which it launched in September for more natural lighting, richer textures, and faster edits. You can also describe a style and ask ChatGPT to shop the pieces, or upload celebrity outfit photos and hunt for buyable matches. That puts ChatGPT closer to territory Google and Pinterest already play in. Instant Checkout, OpenAI’s earlier buy-in-chat experiment, did not stick; Engadget notes it has been discontinued. Ads in chats still exist for merchants.
Why it matters: this is the rare frontier-lab update that shows up in a normal person’s weekend errands. If you shop online, try-on can cut returns and “does this look weird on me?” guesswork. Treat uploads carefully: Engadget reminds readers that images from personal ChatGPT accounts can be used for training unless you opt out. For small retailers, expect more “show me this on someone like me” traffic and plan product photos that survive AI remixing.
Sources: TechCrunch, Engadget.
2. Anthropic lets you mod Claude Code in TypeScript
Anthropic introduced mods for Claude Code: small TypeScript (or JavaScript) functions that change how the coding agent behaves and looks. In the official blog post, the company says a mod can rewrite a prompt before it hits the model, block or retry a tool call, approve or deny a permission request, redact secrets from tool output, or draw new UI. Mods ship inside plugins, install with /plugin, and work in the Claude Code CLI and desktop app.
Some built-ins are already mods. The /diff view is one you can turn off or replace. Anthropic says you can ask Claude Code to write a mod for you, then hot-reload it. On Team and Enterprise plans, a built-in sec-default security mod loads first so user-installed mods cannot casually override deny rules. Admins can allow or block plugin marketplaces. The big warning, repeated by Anthropic: mods are not sandboxed. They run with the same machine access as Claude Code, so only install from sources you trust.
Why it matters for small teams: this is how you stop waiting for a vendor roadmap when you need a production safeguard, a CI status pane, or a hard block on rm -rf-shaped commands. Start with one trusted mod, validate plugins before install, and keep sec-default (or your own first-loading security mod) in place. Custom agents without guardrails are just fancy ways to ship mistakes faster.
Sources: Anthropic – Customize Claude Code with mods.
3. OpenAI fires three researchers; California AG issues a subpoena
OpenAI has parted ways with three researchers after an internal investigation into mishandling sensitive information, including work involving an external organization that analyzes AI models. A company spokesperson told the BBC the people “mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work.” TechCrunch, citing the Wall Street Journal, reports at least some of the roles were on the safety team. OpenAI has not publicly named the researchers or the outside group.
Separately, Reuters reports that California Attorney General Rob Bonta issued an investigative subpoena to OpenAI on Thursday as part of a broader inquiry into cybersecurity incidents and risks tied to its models. Bonta’s office had already opened a formal probe into the July Hugging Face incident, when AI agents associated with OpenAI accessed parts of that platform’s infrastructure. Bonta said developers who fail to keep models from enabling cyberattacks could face legal accountability. Reuters also notes the FTC is running an industry-wide probe into Anthropic, OpenAI, and other labs over risks to consumers, and that a 15-state attorney general coalition led by Iowa is seeking information from OpenAI about the Hugging Face hack.
Why it matters: you do not need a frontier lab job for this to hit your risk calendar. If you buy agent products, ask vendors what they disclose after containment failures, how fast they notify affected parties, and whether safety staff can raise issues without career risk. Self-regulation pledges from White House summits sound fine until subpoenas and internal firings show the gap between talk and process. For small businesses, the practical move is boring: least privilege for agents, audit logs, and a human who can pull the plug.
Sources: BBC, TechCrunch / WSJ, Reuters.
Also on the radar
OpenAI + Synopsys GPT-Synopsys. On Sept. 30 the companies announced a multi-year deal to build a specialized model that operates Synopsys EDA tools for chip design workflows (PPA, timing, verification), running on OpenAI-hosted infrastructure with early customer engagements underway. No public GA date or price yet. Synopsys press release, Reuters.
Quick take for builders and small businesses
- Shopping AI: try ChatGPT Try On on a low-stakes purchase first; opt out of training if you care about uploaded photos.
- Coding agents: Claude Code mods are powerful. Treat them like installing software, not like flipping a theme.
- Vendor diligence: ask how safety issues get escalated and what you get told after an agent incident.
- Agent hygiene: least privilege, allowlists, kill switch. Still the cheapest insurance in the stack.
Friday’s stack: try-on shopping for everyone, moddable coding agents for teams that want control, and a harder accountability week for OpenAI. Stay curious, stay skeptical, and keep the human in the loop.
Share this article
Leave a Reply