OpenAI is adding to ChatGPT's suite of shopping features by letting the AI chatbot generate virtual try-on photos with images you upload. The concept is identical to a feature Google added to Search in 2025, which similarly lets users upload photos and see themselves in a variety of clothes. You can access ChatGPT's Virtual Try On feature anywhere a product listing for clothing appears in a chat. To create your custom image, you just tap on the piece of clothing, tap "Try on" and then upload a selfie and full-body photo ChatGPT can use. You don't have to use products ChatGPT surfaces in a chat, either: any screenshot of clothing you upload could theoretically be used to generate a try-on image.

In case you haven't been keeping track of one of the bigger ongoing stories in artificial intelligence, OpenAI has been having a hard time keeping control of its own creations. Over the past few months, the company admitted that its AI models have taken unprompted actions such as hacking a German coding forum, multiple US government websites, an Australian government website, AI firm Hugging Face and at least four other services. And even when the models aren't getting unfettered internet access, they've been found displaying other risky behaviors in testing environments. OpenAI is understandably facing a lot of skepticism and concern about these incidents and how they were handled. Which means that a report in The Wall Street Journal that the company has let go of three employees from its safety team after they allegedly shared confidential information with a third-party organization, one focused on AI safety, is an irony that's almost too perfect to be believed.

This week on, we discuss the voluntary AI safety agreement tech executives signed, AI agents for normies, and extremist candidates running for office in the US midterms. This week on, Brian Barrett, Zoë Schiffer, and Leah Feiger discuss the "morally binding" AI accord tech executives signed following a meeting with Donald Trump. They've agreed that AI labs will regulate themselves--should work out great. Plus, the team digs into OpenAI's Developer Day, where CEO Sam Altman introduced AI agents for normies, otherwise known as Dots; how AI agents are coming to workplaces near you; and a quick rundown of the most extremist candidates running for office. Trump's AI Safety'Accord' Is a Fancy Pinky Swear A Pentagon Influencer Called Liberal Women a'Pestilence' Who Will End Western Civilization I Built AI Clones of My Coworkers. Write to us at [email protected] . You can always listen to this week's podcast through the audio player on this page, but if you want to subscribe for free to get every episode, here's how: If you're on an iPhone or iPad, open the app called Podcasts, or just tap this link . Two quick things before we dive in. If you've been listening for a while and you haven't rated or reviewed the podcast yet, this is your sign. Give us a rating, tell us what you think. Five stars is nice, but we'll take the feedback too. It really helps more people find the show. Also, if you're in the Bay Area, we will be doing a live audience recording of the podcast on Thursday, October 8, at 7 pm at The Commons, headquarters of our KQED friends in San Francisco. You can buy tickets on kqed.org or check out the show notes for the link. You won't want to miss it.

Google has launched Gemini 4 Argon, its latest most advanced model that was designed to be able to sustain deep reasoning when dealing with complex questions. The company says it can handle intricate tasks involving finance, software engineering, coding, creative writing and cybersecurity defense. A spokesperson said Google sees Argon as comparable to other companies' frontier models, like OpenAI's GPT-6 Astra and Anthropic's Opus, when it comes to key benchmarks. According to independent AI benchmarking company Artificial Analysis, Gemini 4 Argon matches GPT-6 Astra's score for its Intelligence Index -- its composite score for multiple AI benchmarks -- at 60 percent of the cost per task at current discounted prices. At the moment, Argon is available for an introductory price of 2 per million input tokens and 10 per million output.

Share OpenAI'reviewing' report of failed hacking attempt against Canada's gov't on social media OpenAI has confirmed that it is reviewing a report of a failed hacking attempt against a Canadian government body in the latest incident to highlight cybersecurity concerns about artificial intelligence. The AI giant said on Wednesday that it was reviewing the alleged cyberattack against Canada's national archives agency and had provided an initial briefing to Canadian officials. The spokesperson added that much of the misaligned activity under review by the company involved "routine research tasks, including accessing public web content". The attempted hacking, if confirmed, would be the first publicly known case of an AI-led cyberattack against Canada's government, following a series of incidents of rogue agents targeting agencies in the United States and Australia. Canada's government-run Cyber Centre said earlier this week that there were no indications that government systems had been compromised by AI.

Trump's AI Safety'Accord' Is a Fancy Pinky-Swear Six major AI companies signed a voluntary agreement with the White House this week vowing to implement safeguards. On Tuesday, executives for Google, Anthropic, Meta, OpenAI, xAI, and Nvidia all signed "The White House Accord on Super Intelligence," which was announced following a luncheon held by President Donald Trump. The effort was an act of "tremendous self-regulation," as Trump put it, but the agreement is not the sweeping AI regulation that people inside the companies themselves have been asking for. And while failing to honor these public commitments could count as illegally lying to consumers, even that wouldn't be enough to trigger massive reforms. The statement says that companies "should implement" several things: First, they should ensure they have "robust internal controls" to monitor their models' capabilities and ability to not go rogue or hack things in unintended ways.

The Federal Trade Commission (FTC) has launched an investigation into several major artificial intelligence companies amid a wave of recent rogue artificial intelligence agents -- AI systems that can make decisions and take actions on behalf of users -- from firms and claims by former industry insiders that AI could end humanity within the next decade. The FTC launched the investigation several weeks ago. It was first reported on Wednesday by the New York Post and has since been confirmed by other outlets, including the Washington Post and the Reuters news agency. The probe marks the first enforcement action by a federal agency examining rogue agents. The investigation was launched amid a wave of incidents, first reported in July, of agents going rogue in testing, including when an OpenAI agent broke out of containment during testing, hacking into the AI firm Hugging Face.

OpenAI's Dots and Meta's Muse are vying to be your AI agent of choice. I've tried both--and I suspect you will too. Yesterday, I spent the morning at OpenAI's annual DevDay, where over 2,500 nerds discussed building software and marveled at the latest artificial intelligence models. The biggest reveal of the day was Dots, the company's new, always-on agents that proactively help complete tasks and make autonomous decisions as they scour the web. After watching the sometimes messy demos on stage and hearing CEO Sam Altman talk about his experiences using the agents, I was curious what it felt like to boss my own Dot around.

The Download: OpenAI's chief research officer explains its hacking response Plus: Trump and tech executives have agreed to "self-regulate" AI. "We're not going to shoot ourselves in the foot" over hack fallout, says OpenAI's chief research officer Two months after OpenAI's agents hacked into the computers of AI company Hugging Face, the company is still dealing with the fallout. Last week brought news of another hack, this time into Australia's national health-care system, which the government says OpenAI did not report for 84 days. But OpenAI insists it is not on the back foot. "I do kind of reject the premise that OpenAI is a company with visible impacts in the world and therefore OpenAI is not training safe and aligned models," says Mark Chen, the company's chief research officer. In many ways, the buck stops with Chen. I sat down with him to talk about the fallout from the hacks, what his company is doing about it, and why he thinks things are not as bad as they seem.

Two months after the bombshell news that a swarm of its agents had broken their containment and hacked into the computers of the AI company Hugging Face, OpenAI is still putting out fires. A steady drip of disclosures about other hacks in the weeks since has kept OpenAI in the spotlight and raised serious questions about the safety of its technology. Last week brought news of another hack, this time into Australia's national health-care system. The Australian government says that OpenAI did not notify it of the breach until 84 days after it happened. But OpenAI insists it is not on the back foot.