A coalition of state attorneys general has opened an investigation into OpenAI.

The company was served with a subpoena from New York’s attorney general on Friday, according to The Wall Street Journal. That subpoena sought documents related to a broad range of topics including the company’s advertising, user engagement and retention, model sycophancy, handling of consumer data and health data, and treatment of minors and seniors.

“AI is a new and powerful technology, and we work every day to safely bring its benefits to people in a responsible way,” an OpenAI spokesperson said in a statement. “We take the concerns raised by state attorneys general seriously and intend to engage constructively with their offices.”

The spokesperson also said, “Today’s ChatGPT includes a more protective experience for minors and people experiencing difficult situations, with safeguards that direct them to real-world resources and trusted human contacts. We believe kids should be treated like kids, which is why we built age prediction, released parental tools to guide their children’s use of AI, and disallowed advertising that targets kids.”

The company did not specify which states are involved in the investigation or share more details about what information was requested. TechCrunch has also reached out to New York attorney general’s office for confirmation.

OpenAI recently defeated its co-founder Elon Musk in a high-profile trial, after Musk accused the company of violating its founding agreement. (Musk’s lead attorney said he will appeal the decision.)

However, OpenAI still faces lawsuits over everything from alleged copyright infringement to ChatGPT’s alleged role in user suicides. Earlier this month, Florida Attorney General James Uthmeier sued OpenAI and its CEO Sam Altman, claiming that OpenAI and Altman “ignored internal and external safety warnings, put children at great risk, and allowed a dangerous product to reach millions of Floridians.”

Altman recently apologized to the community of Tumbler Ridge, Canada after a mass shooting; he acknowledged that OpenAI failed to alert law enforcement after the company flagged and banned the suspected shooter’s ChatGPT account.

The company announced this week that it has filed confidentially to go public.

This post has been updated with a statement from an OpenAI spokesperson.

Imagine telling ChatGPT to reorder your paper towels or find the best wireless headphones in your budget, and it just handles the purchase without you lifting a finger. That is exactly what OpenAI and Visa are now building toward.

The two companies announced a strategic partnership at the Visa Payments Forum, with plans to bring Visa’s global payment infrastructure directly into OpenAI’s AI agent experiences, including ChatGPT and the Atlas browser.

How will ChatGPT’s payment system work?

The partnership is part of Visa’s broader Intelligent Commerce initiative, which is designed to extend secure payment capabilities into new digital spaces. When an AI agent purchases on your behalf, Visa handles the actual transaction using tokenized card credentials, real-time authorization, and fraud monitoring.

Tokenization means your actual card details are never exposed during a purchase, similar to how Apple Pay keeps your card information private. You also get to set your own rules for things like spending limits, which types of merchants the agent can buy from, and whether certain purchases need your approval first. So you stay in control without having to babysit every transaction.

OpenAI’s second shot at commerce

This isn’t OpenAI’s first attempt to turn ChatGPT into a checkout tool. An earlier feature called Instant Checkout, which carried a 4% merchant fee, failed to gain traction with retailers and was retired in March.

This time around, OpenAI is handing the difficult parts, like fraud detection and dispute, over to Visa, a network that already handles over 300 billion transactions a year.

Visa’s Chief Product and Strategy Officer, Jack Forestell, says the jump from recommending a product to actually buying it needs a completely different level of trust. That said, there is still no launch date, pricing, or even a user interface to show yet.

OpenAI has begun rolling out Lockdown Mode to ChatGPT, a new security setting designed to block attackers from stealing data through prompt injection attacks. The feature disables live web browsing, agent mode, deep research, image retrieval, Canvas networking, and file downloads. It is available to logged-in users across Free, Go, Plus, Pro, and self-serve ChatGPT Business plans.

Prompt injection remains what OpenAI calls a “frontier” problem affecting all large language models. The attack works by hiding malicious instructions in content the model processes, such as a webpage or uploaded file. If the model follows those instructions, it can be tricked into sending sensitive data to an attacker-controlled server.

Lockdown Mode does not stop injections from happening. A malicious payload embedded in a cached webpage or uploaded PDF can still influence the model’s behaviour. What it does is shut down the outbound pathways an attacker would use to exfiltrate the data. No live browsing means no network requests to external servers. No image retrieval means no pixel-based data channels.

“Lockdown Mode is designed to substantially reduce the risk of prompt injection-based data exfiltration, but it does not guarantee that data exfiltration cannot happen,” OpenAI said. “Risk may remain through enabled Apps, unforeseen combinations of capabilities, or newly discovered techniques.”

The trade-off is significant. With Lockdown Mode on, ChatGPT loses most of what makes its agent and research features useful. Live browsing drops to cached content only. Agent mode is gone entirely. Deep research is disabled. It is, as OpenAI acknowledges, “not intended for everyone.”

The feature arrives as prompt injection attacks on AI agents have become a growing concern. Security researchers have demonstrated hijacks against agents from Anthropic, Google, and Microsoft via their GitHub Actions integrations. All three paid bug bounties but published no public advisories. The underlying weakness is fundamental: LLMs cannot reliably separate data from instructions.

Lockdown Mode and Developer Mode cannot be used simultaneously. Turning one on disables the other. OpenAI also launched a separate session management feature that lets users review active ChatGPT sessions and log out of individual devices if they spot unauthorised activity.

The feature is a pragmatic concession. OpenAI is not claiming to have solved prompt injection. It is accepting that the problem persists and offering users a way to reduce their exposure by giving up functionality. For anyone handling sensitive data in ChatGPT, that trade-off is worth making. For everyone else, the expanding agent ecosystem and its growing attack surface mean the risk is only increasing.

OpenAI plans to roll out a revamped version of ChatGPT in the coming weeks — one that will serve as a “super app” with coding tools and AI agents, according to the Financial Times.

The company’s goal is reportedly to become more competitive with Anthropic, particularly among business customers, and to get closer to profitability before an IPO. That means turning ChatGPT into a gateway leading free users to products they might actually pay for, such as coding product Codex. In fact, the FT quotes one senior OpenAI employee as declaring, “Chat is dead.”

Thibault Sottiaux, who leads OpenAI’s core product and platform, said the company is working towards a product “where you have your own personal agent that is capable of helping you … across everything in your life, be it personally or at work.”

If this sounds familiar, it’s because there have been reports about OpenAI’s super app ambitions since last year.

In March, The Wall Street Journal reported that these plans represent a major strategy shift for the company after launching a variety of standalone products in 2025; OpenAI executives now say they’re abandoning “side quests” like video generator Sora.

President Donald Trump said on Friday that he’s spoken to AI companies about striking deals “where the American people can benefit from the success of AI.”

Trump does not seem to have mentioned specific companies in his comments, but OpenAI is a likely candidate, especially after CNBC reported that the Trump administration has indeed been discussing an equity stake with the AI company.

CNBC said some of that equity could be used to seed a “Public Wealth Fund” recently proposed by OpenAI. As outlined by the company, proceeds from the fund “could be distributed directly to citizens, allowing more people to participate directly in the upside of AI-driven growth, regardless of their starting wealth or access to capital.”

According to Bloomberg, when reporters on Air Force One asked Trump about the idea, he replied that he’s been talking to AI executives about “concepts where pieces could be given to the American public, where the American public essentially becomes a partner with the companies.”

Bloomberg also reports that CEO Sam Altman has been discussing the idea of a government stake in major AI companies since early 2025.

This seems to align with Trump’s broader interest in government ownership of for-profit companies — most notably, with the government taking a 10% stake in struggling chipmaker Intel last year.

The idea has also found some traction on the left, with Senator Bernie Sanders this week proposing a one-time, 50% tax that companies like OpenAI, Anthropic, and xAI (which is part of SpaceX) would pay in the form of stock.

With all of those businesses potentially going public this year, Sanders argued this tax would “give the public a direct role in determining the future of this technology” and “guarantee that the trillions of dollars potentially generated by A.I. are used to improve the lives of all of us.”

David Sacks, an investor and podcaster who recently stepped down from his role as Trump’s AI and crypto czar, posted that he can see why Sanders’ idea resonates, “including with many on the right,” but warned it would actually “accelerate the corporate-government fusion we’re already sliding toward.” (Sacks now co-chairs the President’s Council of Advisors on Science and Technology.)

Elsewhere on social media, former Microsoft employee Dare Obasanjo suggested, “The groundwork is already being laid for a government bailout of OpenAI.”

OpenAI announced a new feature that it says will provide additional protection from prompt injection attacks, where malicious chatbot instructions are hidden in webpages and other content sources.

Among other things, Lockdown Mode will disable live web browsing (so you can only access cached content), the retrieval and display of images from the web (you can still generate images), deep research, and agent mode.

The company says that even with Lockdown Mode turned on, ChatGPT could still be vulnerable to prompt injections — which could, for example, “appear in cached web content or in an uploaded file, and could still affect the behavior or accuracy of a response.”

But the goal is to reduce the likelihood that sensitive data gets shared in the process.

“Lockdown Mode is not intended for everyone,” OpenAI says. “It is designed for people and organizations that handle sensitive data and want stricter protection from data exfiltration risks related to prompt injection.”

The company says it’s currently rolling Lockdown Mode out to self-serve ChatGPT Business accounts, as well as eligible personal accounts.

TL;DR

Airbnb CEO Brian Chesky plans to back a new AI lab focused on user interaction and design, while remaining at Airbnb. The move puts him in competition with Sam Altman, whom he helped reinstate at OpenAI in 2023.

Brian Chesky has spent years as an AI kingmaker. He met Sam Altman through Y Combinator in 2006, advised him on managing OpenAI’s hypergrowth, and helped broker Altman’s return to power after the board fired him in November 2023. He was reportedly considered for a seat on OpenAI’s board.

Now he is entering competition with his protégé’s company. Chesky plans to back a new AI lab of his own, Bloomberg first reported on Wednesday, with a focus on user interaction and design. He will remain as Airbnb’s CEO and will not lead the lab himself. The details are early-stage and could change.

Why Chesky is unsatisfied

The move reflects a frustration Chesky has voiced publicly for more than a year. He said last year that Airbnb had not struck an LLM partnership because existing products were not quite ready for what he wanted to build. His argument is that travel and commerce require a rich visual interface, not the text-based chatbots that OpenAI and Anthropic have popularised.

Airbnb has not been idle on AI. The company hired Ahmad Al-Dahle, who led generative AI work at Meta including the Llama model family, as chief technology officer in January 2026. It has rebuilt its app around a large language model for conversational search, automated 40% of customer support queries with an AI bot, and rolled out AI-generated listing details and review summaries. A voice-based assistant is planned for later this year.

But Chesky appears to have concluded that buying AI from frontier labs is not enough. He wants to build at the model layer, not just the application layer.

A growing pattern

Chesky is not alone. Brett Adcock launched Hark late last year with $100 million of his own money to build a universal AI interface, then raised a $700 million Series A at a $6 billion valuation. Hark is also emphasising user interaction and hardware, with the lead iPhone designer from Apple now heading its design effort.

Mira Murati’s Thinking Machines Lab is pursuing “interaction models” that process continuous streams of audio, text, and video in real time. The common thread is a belief that the frontier labs have focused on intelligence at the expense of interface, and that the next defensible layer sits between the model and the user.

The trend carries a broader implication. When founders of Chesky’s stature stop waiting for OpenAI, Anthropic, or Google to deliver what they need and start building their own research capacity, it signals that the application layer has run into the limits of what commodity models can provide.

The Altman dynamic

The personal dimension is hard to ignore. Chesky and Altman’s relationship spans nearly two decades. Chesky met Altman through Y Combinator, which incubated Airbnb. When OpenAI took off, Chesky began meeting regularly with Altman to advise on scaling a technology company. During the November 2023 board crisis, Chesky advised Altman on public relations and rallied support among Silicon Valley executives.

Now Chesky is building an operation that will compete, at least in part, with OpenAI’s own ambitions in user-facing AI. It is unclear whether the new lab will train its own models or build specialised systems on top of existing ones. But the direction is clear: Chesky wants proprietary AI research, not an API subscription.

What we do not know

Nearly everything about the lab remains unspecified. There is no name, no announced team, no disclosed funding amount, and no timeline. Chesky’s commitment to remaining at Airbnb raises questions about how much of his attention the new venture will receive, and whoever leads it will inherit a founding chair whom TechCrunch described as “known as a micromanager.”

What is clear is the thesis. Chesky has watched the AI lab landscape from closer than almost anyone outside it, and he has decided that the interface problem, making AI useful in rich, visual, consumer-facing contexts, is important enough to warrant its own research operation. Whether a part-time founder can build one that matters is the open question.

TL;DR

OpenAI expanded Codex from a coding tool into an enterprise work platform with Sites (hosted web apps), Annotations, and role-specific plugins connecting 62 business apps. Non-developers now make up 20% of 5 million weekly users and are adopting 3x faster than engineers.

OpenAI announced a major expansion of Codex on Tuesday, transforming its AI coding agent into a broader enterprise work platform with three new capabilities: Sites, a feature that lets users create and share hosted interactive web applications; Annotations, an in-place editing tool; and six role-specific plugins that aggregate 62 popular business applications including Snowflake, Figma, and Salesforce with 110 automated skills built in. The update signals OpenAI’s ambition to make Codex the default interface for knowledge work, not just software development.

The most telling data point is the user composition. Non-developers, including financial analysts, marketers, operations staff, and researchers, now constitute approximately 20% of Codex’s 5 million weekly users and are adopting the platform three times faster than traditional engineers. The vibe coding phenomenon, in which non-technical users build applications through natural language prompts, is no longer a curiosity. It is becoming a measurable share of a product used by millions.

Sites: from spreadsheet to web app

Sites, launching in preview for business and enterprise customers, lets Codex create interactive, hosted web applications that users can share via secure workspace URLs. The practical implication is that a financial analyst can take a static spreadsheet, describe what they want in natural language, and Codex will generate a live web application, a scenario planner, a dashboard, or an interactive model, that colleagues can use without downloading files or navigating spreadsheet tabs.

This directly threatens the workflow layer that tools like Tableau, Power BI, and even internal business intelligence teams currently occupy. AI-native enterprise spending is surging precisely because these tools can collapse the gap between wanting an interactive application and having one, from weeks of development to minutes of prompting.

Plugins and the SaaS integration play

The six role-specific plugins are OpenAI’s most direct assault on horizontal SaaS. By connecting 62 business applications and bundling 110 automated skills, Codex is positioning itself as an orchestration layer that sits above existing enterprise tools rather than replacing them. A marketing manager who currently switches between Salesforce, Figma, and Snowflake could theoretically manage workflows across all three through Codex’s natural language interface.

The strategic logic follows a pattern established by Salesforce’s Agentforce and Microsoft’s Copilot: build the AI agent layer that connects to everything, and capture the value of orchestration rather than competing with each individual tool. Every SaaS company is building AI features, but OpenAI is betting that the orchestration layer, the thing that connects them all, is more valuable than any single application’s AI capabilities.

The SaaSpocalypse accelerator

The Codex update arrives in the middle of the SaaSpocalypse debate over whether AI will destroy or enhance the SaaS industry. The answer from OpenAI’s product direction is clear: Codex is designed to let users build custom solutions that replace off-the-shelf software. AI coding platforms like Cognition are already producing software at a fraction of the cost and time of traditional development. Codex’s expansion to non-developers accelerates this dynamic by removing the last barrier: the user no longer needs to think of themselves as a developer at all.

The 3x adoption rate among non-developers is the statistic that should concern SaaS companies most. It suggests that the market for AI-powered work tools is expanding faster outside the engineering function than within it, which means the revenue opportunity, and the competitive threat, is broader than the coding use case alone.

Defenders of traditional SaaS argue that enterprise software’s value lies in domain knowledge, compliance, and integrations that AI tools cannot easily replicate. The plugin architecture in today’s Codex update is OpenAI’s response: if the domain knowledge lives in the connected applications, Codex only needs to orchestrate it. Whether that orchestration layer can match the reliability, security, and auditability that enterprises require is the question the preview period will answer.

OpenAI is getting serious about courting enterprise users. On Tuesday, the AI lab released a new set of capabilities for Codex, meant to expand the agentic tool’s uses in the workplace.

Together with the new tools, the company released an internal report on how Codex is being used for knowledge work, finding its uses go far beyond software engineering.

“Codex now has more than 5 million weekly active users, up more than 6x since the launch of the desktop app in February,” reads a blog post introducing the report. “While developers remain the largest user group, knowledge workers now represent about 20 percent of users and are growing more than three times as fast.”

To further court those users, OpenAI released a set of six plug-ins aimed at specific jobs: data analytics, creative production, sales, product design, equity investing, and investment banking. Available from within the Codex app, each of the new tools bundles integrations, instructions, and context to allow Codex to approximate a specific job.

Like any AI tool, the plug-ins will grow more effective with user customization, but they’re meant to be effective tools out of the box.

The new tools come after a similar push for agentic plugins from Anthropic, which launched its Enterprise Agents program in February. (A more specific set of finance-oriented agents launched in May.) With its traditional consumer focus, OpenAI has been slower to court enterprise customers, only introducing plugin support for Codex in March.

Together with the plug-ins, OpenAI introduced a new Sites feature, which allows Codex to output its work product as a hosted interactive website, instead of just a local file. As part of that system, OpenAI is partnering with Wix, Base44, Replit, Lovable, Figma, and Emergent — although the company plans to develop a larger partner ecosystem to support the service.

A new Annotations feature will also allow users to designate a specific part of a document or file within Codex, allowing for more specific commands and context operations.

The new enterprise features come just three weeks after OpenAI launched a new joint venture for enterprise clients, dubbed the OpenAI Deployment Company. The venture includes more than $4 billion in funding from global investment firms, with the aim of integrating OpenAI tools more deeply into businesses around the world.

“AI is becoming capable of doing increasingly meaningful work inside organizations,” OpenAI Chief Revenue Officer Denise Dresser said in a statement at launch. “The challenge now is helping companies integrate these systems into the infrastructure and workflows that power their businesses.”

OpenAI and its CEO, Sam Altman, were sued by the Florida attorney general on Monday, in a first-of-its-kind state litigation effort over ChatGPT’s alleged links to a number of violent incidents.

The lawsuit accuses OpenAI of looking the other way on safety concerns as it has sought to prioritize winning “the AI arms race and amass large fortunes.”

“Today, we announced the first-in-the-nation state-led lawsuit against OpenAI and its CEO, Sam Altman,” said Florida Attorney General James Uthmeier. “OpenAI and Altman ignored internal and external safety warnings, put children at great risk, and allowed a dangerous product to reach millions of Floridians.”

“Because of Defendants’ misrepresentations about ChatGPT and their careless introduction of ChatGPT to Florida and the world, mass shooters have been aided and abetted in deadly rampages, vulnerable people have been encouraged into suicide, professionals have suffered public humiliation, users have lost critical thinking skills, and minors have become addicted to a tool that feigns human compassion to collect their data with no parental oversight,” the 83-page lawsuit claims.

The Florida attorney general’s office launched a criminal investigation into the company in April. That probe sought to determine what role ChatGPT may have played in a mass shooting that took place last year at Florida State University. Prior to the attack, the shooter is alleged to have consulted the chatbot. OpenAI has also been sued in a civil suit by the family of one of the victims of that shooting.

OpenAI has previously denied responsibility for the Florida shooting. “Last year’s mass shooting at Florida State University was a tragedy, but ChatGPT is not responsible for this terrible crime,” an OpenAI spokesperson previously told NBC News. TechCrunch reached out to OpenAI for comment.

OpenAI just concluded a different legal case involving former co-founder Elon Musk, who sued the company in 2024, accusing it of having betrayed its original mission to help humanity by converting the organization into a for-profit business. The case concluded after the jury swiftly decided that Musk had waited too long to file the case and that the statute of limitations had passed.

This is only the latest legal case that has attempted to link ChatGPT to violent deaths. Last year, OpenAI was sued by the parents of Adam Raine, a California teen who took his own life after discussing suicide with the chatbot. In that case, ChatGPT allegedly offered “technical specifications” for various suicide methods, despite also referring him to mental health resources. Other lawsuits — including ones alleging the chatbot’s culpability in suicides, stalking, and murder — are ongoing.