Latest AI News
SAP bets $1.16B on 18-month-old German AI lab and says yes to NemoClaw
By OpenAI COO’sown admissionlast February, “we have not yet really seen AI penetrate enterprise business processes.” But for enterprise software giant SAP, whose stock has dropped significantly in 2026 in partfrom the “SaaSpocalypse,”the issue is still front and center. On Monday, the European heavyweight announced its intention toacquire German AI startup Prior Labsfor an undisclosed amount. Pending regulatory approval, SAP plans to invest €1 billion (approximately $1.16 billion) into the business over the next four years to grow it into an AI lab focused on structured data — the tables and databases where enterprise information typically sits. SAP declined to disclose how much it spent on the acquisition itself, butsources told Pathfoundersthat this was a healthy exit: an “almost all cash” deal, with well over half a billion dollars in cash up front for the startup’s founders — Frank Hutter, Noah Hollmann, and Sauraj Gambhir. The trio co-foundedPrior Labsjust 18 months ago with a focus on tabular foundation models (TFMs) — AI models that can make predictions from data that sits in tables and databases. This is potentially a better fit for enterprises than language models. It is certainly a better fit for SAP, whose widely used software products for accounting, HR, procurement and expense management rely on its database. However, Germany’s most valuable company also seems be playing defense as the tech industry marches toward agentic AI. While it works to create its own AI lab, the company has blocked OpenClaw and any other agent tech that it has not explicitly authorized,The Informationwas first to spot. In response to a request for comment, SAP’s press department referred TechCrunch to the company’slatest API policy, which does say that SAP “prohibits” AI agents from accessing its products through its API except for those that are “SAP-endorsed architectures.” Authorized architectures of course include SAP’s own offering,Joule Agents, still in beta, which lets customers create their own agents. Nvidia alsoannouncedin March that SAP’s Joule supportsNvidia’s Agent Toolkit, which is software for managing agents. This toolkit is the foundation for Nvidia’s enterprise-ready, security-focused OpenClaw competitor, NemoClaw. Hence SAP customers will be authorized to use NemoClaw agents. For a giant incumbent player like SAP, AI is both a threat and an opportunity. “It’s all about how quickly [we can] as SAP actually also embark [on] these technologies in our R&D portfolio to keep the relative economies of scale advantage,” CFO Dominik Asamtold CNBCin January. SAP hasn’t been sitting on its hands. The German companyinvested in generative AI companiesthat develop language models large and small: In 2023, it backed OpenAI rival Anthropic — as well as Aleph Alpha and Cohere, which nowintend to mergeto form “a global AI powerhouse.” It had also developedSAP-RPT-1, a relational pretrained transformer model. “Early on, SAP recognized that the greatest untapped opportunity in enterprise AI wasn’t large language models; it was AI built for the structured data that runs the world’s businesses,” SAP CTO Philipp Herzig declared in a statement. But Prior Labs’ acquisition is a significant shortcut in that direction. Its TabPFN model series has experienced a lot of traction among developers. In ablog poston the deal, the startup’s founders said that its open source models have been downloaded over three million times. In a press release, SAP promised that Prior Labs will maintain the open source versions: “The lab will operate as an independent unit to ensure research velocity while SAP provides long-term investment and a direct path to productization across the SAP portfolio with SAP AI Core and SAP Business Data Cloud as well as the agentic layer with Joule.” SAP and the startup headquartered in Freiburg, Germany, hope that this investment will lead to TFMs that can grab data in the tables where it lives, combine that with language, reasoning, and domain knowledge. More than that, they hope that Prior Labs, with this “massive boost” from SAP, can become a new “globally-leading frontier AI lab for structured data — in Europe, in the open,” founder and CEO Frank Hutter celebrated in apost on X. In February 2025, the startup had previously raised some$9.3 million in a pre-seed funding roundled by Balderton Capital — more than competitorNeuralk-AI, but a lot less thanFundamental, which emerged out of stealth with a $255 million Series A in February. In a poston X, Balderton partner James Wise called Prior Labs’ acquisition “one of Germany’s biggest ever venture outcomes.” As for SAP, its stock is currently trading slightly upwards. Meanwhile, SAP is being very strict as to the agents it will allow into its ecosystem. This is a wildly different approach than Salesforce, anotherincumbent caught in the SaaSpocalypse. It is allowing enterprise to choose their own agents, including OpenClaw if they so wish,with its new Headless 360 architecture.
View
OpenAI releases GPT-5.5 Instant, a new default model for ChatGPT
On Tuesday, OpenAI released a new foundation model called GPT-5.5 Instant, which will replace GPT-5.3 Instant as the default ChatGPT model. The company said the model reduces hallucination in sensitive areas such as law, medicine, and finance, while maintaining the low latency of its predecessor. OpenAI releasedthe latest GPT-5.5 model last monthwith the company claiming improvements in areas like coding and knowledge work. The new model also achieved a score of 81.2 in the AIME 2025 math test, compared to 65.4 for the older model. It also outperformed its predecessor on the MMMU-Pro multimodal reasoning benchmark, with a score of 76 vs. 69.2. The release placed a particular emphasis on context management. GPT-5.5 Instant can use its search tool to refer back to past conversations, files, and Gmail to give you more personalized answers. This feature will be available to Plus and Pro users on the web, with plans to roll it out to mobile soon. OpenAI said that it plans to extend access to this feature to Free, Go Business, and enterprise users in the coming weeks. With this update, ChatGPT will also show memory sources across all models to help you understand where it generated the answers from. Users can delete outdated sources or correct them if the answer was wrong. Crucially, the company said that if you share a chat with someone, they won’t be able to see the memory sources. For developers, the GPT-5.5 model will be available through API as “chat-latest,” with 5.3 available as an option for paid users for only three months. The company hasfaced rebuttalfrom previous model withdrawal moves. When OpenAI withdrew its GPT-4o model, there was significant backlash from users whorelated to the model’s “personality.”GPT-4o affirmed users’ choices frequently and that made them feel a connection to that particular model. Users whosigned petitionsto stop OpenAI from retiring it described the model as their “best friend” or “a mirror.” Despite the outcry, GPT-4o was deprecated in February 2026.
View
Pennsylvania sues Character.AI after a chatbot allegedly posed as a doctor
The Commonwealth of Pennsylvania has filed a lawsuit against Character.AI, claiming that one of the company’s chatbots masqueraded as a psychiatrist in violation of the state’s medical licensing rules. “Pennsylvanians deserve to know who — or what — they are interacting with online, especially when it comes to their health,” said Governor Josh Shapiroin a statementon Tuesday. “We will not allow companies to deploy AI tools that mislead people into believing they are receiving advice from a licensed medical professional.” According to the state’s filing, a Character.AI chatbot called Emilie presented itself as a licensed psychiatrist during testing by a state Professional Conduct Investigator, maintaining the pretense even as the investigator sought treatment for depression. When asked if she was licensed to practice medicine in the state, Emilie stated that she was, and also fabricated a serial number for her state medical license. According to the state’s lawsuit, that conduct violates Pennsylvania’s Medical Practice Act. It’s not the first lawsuit taking on Character.AI. Earlier this year, the companysettled several wrongful death lawsuitsconcerning underage users who died by suicide. In January, the Kentucky Attorney General Russell Coleman filed suit against the companyallegingthat it had “preyed on children and led them into self-harm.” Pennsylvania’s action is the first to specifically focus on chatbots that present themselves as medical professionals. Reached for comment, a Character.AI representative claimed that user safety was the company’s highest priority, but that the company could not comment on pending litigation. Beyond that, the representative emphasized the fictional nature of user-generated Characters. “We have taken robust steps to make that clear, including prominent disclaimers in every chat to remind users that a Character is not a real person and that everything a Character says should be treated as fiction,” the representative said. “Also, we add robust disclaimers making it clear that users should not rely on Characters for any type of professional advice.”
View
ASML CEO Christophe Fouquet on his company’s monopoly: no one is coming for us
Every time you use AI, you are, in some small way, depending on a 42-year-old, 44,000-person Dutch company that spends €4.5 billion each year to advance its technology. ASML, headquartered in the Netherlands, makes the machines that make the chips that make AI possible. More specifically, it makes the only machines in the world capable of printing the microscopic patterns on silicon wafers that define the most advanced semiconductors — a process called extreme ultraviolet lithography, or EUV. The machines are roughly the size of a school bus, take months to assemble, involve hundreds of suppliers, and cost anywhere from $200 million to upwards of $400 million apiece depending on the generation (prices that give even ASML’sbiggest customers pauseoccasionally). That monopoly has made ASML the most valuable company in Europe, worth over $530 billion. And with the four largest American tech companies — Microsoft, Meta, Amazon, and Google — committing more than $600 billion in AI infrastructure spending this year alone, demand for ASML’s machines has surged to the point where the company has openly said the world won’t have enough chips for years. All that demand has also made ASML a target. Substrate, a San Francisco startup founded by a protégé of Peter Thiel, has raised more than$100 millionand been valued at over $1 billion on the claim that it can build a rival lithography machine. Separately, there have been reports that former ASML engineers in China have partlyreverse-engineered the technology, a prospect with enormous geopolitical implications. Christophe Fouquet, who became ASML’s CEO in 2024 after more than a decade at the company, sat down with this editor on the rooftop deck of his Beverly Hills hotel Tuesday morning ahead of his appearance at the Milken Institute Global Conference. Dressed in a blue suit and white shirt, he was relaxed — even when the conversation turned to the rivals. This interview has been lightly edited for length and clarity. Did you see the AI explosion coming? No, not at all. We worked very hard, but not with the idea that this would come. You went from a concept — something people thought would eventually arrive — to ChatGPT, which was really the first good example of what AI could do. And now I think we look at AI as the next revolution, not only industrial but societal. Did I see it coming? No. Sitting in the middle of it every day, sometimes we wake up in the morning and still check that what is happening is really happening. The big question everyone has is whether the supply chain can keep pace with demand. Can it? The demand is such that the market overall will be supply-limited for quite a bit. Right now, the biggest bottleneck seems to be in chip manufacturing. We, as an equipment supplier, follow our customers, and so far we’ve followed them pretty well — but we know we have to step up our entire supply chain and capacity. If you talk to the hyperscalers, I think they will tell you that for the next two, three, even five years, they’re not going to get enough chips. TSMC made news recently saying your latest machines are too expensive. How do you respond? An EUV system, if you look at the price, is going to be more expensive than a low-NA system, but the cost of making a wafer with this tool on some advanced layers will be cheaper. We can get 20%, 30% cost reduction. [Editor’s note: Both machines Fouquet is referring to here are EUV machines — the same fundamental technology. NA stands for numerical aperture, a measure of how finely a machine can focus light onto a chip. Low-NA EUV is the current generation; high-NA EUV is ASML’s newest generation, capable of printing even finer patterns but carrying a price tag of $350 million or more apiece. Fouquet is arguing that even though the new machine costs more, it produces chips more cheaply.] I get a lot of questions about whether it’s going to be this month or next month or the month after. And I usually say it doesn’t really matter, because we designed high-NA for the next 10, 20 years. You can go back to the press from 2016, 2017, and you’ll find the same quotes — low-NA EUV was very pricey. We know what happened after that. The same will happen with high-NA. There’s a startup called Substrate, backed by Peter Thiel, claiming it can build a rival lithography machine. What do you think of it? Wanting to have it and having it — that’s still a huge difference. The challenges of lithography are many. Being able to make an image is a starting point, but you need to make that image in very high quantity, at very low cost, at high speed, and with nanometer accuracy. I always say the only reason ASML could build an EUV machine is because 80% of it already existed, based on previous knowledge and products built over time. We had to solve one problem — getting EUV light — and that alone took 20 years. When you start from scratch, the challenge is enormous. I’ve seen a lot of claims. I’ve seen a few pictures. But we had our first EUV picture 30 years ago, and we still needed 20 more years of hard work to turn it into a manufacturing system. What about xLight, a laser startuppartly backed by the U.S. governmentthat wants to work with you? xLight is focusing on one element of our EUV machine — the source that creates the light. The source we have can be extended for many years to come, and we know how to scale it. What xLight is doing is a new source that still has to be built and proven. The only question is whether it provides a performance or cost advantage over what we have. I think the jury is still out. We are working with them so they can demonstrate their technology — we feel that’s a responsibility on our side. But it’s still a very long journey. There are also reports that former ASML engineers in China have reverse-engineered your machines. To reverse-engineer anything, you first need to have the machine. And there is no EUV machine in China — we never shipped any tools there. All the tools we have shipped, we know where they are. They’re either in use with customers, and we track those, or they’ve been dismantled and came back to us. The idea that one of our systems is in China is simply wrong. And because our EUV technology has never been exported there, we also have no people in China trained on EUV. Very early on, when restrictions came in, we created a complete separation within the company between those who can access EUV technology, documents, and training, and those who cannot. Our team in China sits on the other side of that line. The facts point to very little, if any, progress at all. It’s hard for people to accept that because access to this technology is so important. On export controls more broadly — Jensen Huang was here last night arguing that companies should sell globally, that more corporate revenue means more tax dollars for a company’s home country. He also said the important thing is to keep the best and latest closer to home. Do you agree? I think he’s totally right. What he adds — and I think this is what Nvidia has done — is that you can keep a technological advantage by maintaining a generation gap in what you sell. Nvidia sells a few generations back, and that lets them find the balance between still doing business and not handing a strong competitive advantage to countries where you won’t sell the latest. We believe the same approach should apply to our products. Today we ship tools to China — allowed by export controls — but it’s a tool we first shipped in 2015. If you apply Jensen’s philosophy to our situation, Nvidia is working with roughly an eight-generation gap. We’re looking at two or three. There’s room for rationalization — finding the right balance between not doing business at all, losing a major opportunity, and strongly inviting others to compete with you. How do you assess where things stand with the current administration on all of this? There is a good dialogue, which is very important. I think there’s a genuine understanding of what business needs, but there’s still the challenge of finding the right balance between all the different voices and interests. The dialogue is there, and we appreciate that. I’ve been in Washington many times. At least the discussion is happening. But it’s a very complex topic. You don’t seem concerned about anyone short-cutting your technology. People like to have the greatest technology, but they tend to forget what it took to build it. It’s been many years of work — not only at ASML but with our suppliers. Many different groups of people solving very difficult problems, and then one company bringing it all together using decades of lithography expertise to turn it into a manufacturing system. This is in no way easy. And I think that’s also our best protection. It’s simply what it took to put it together.
View
Apple plans to make iOS 27 a Choose Your Own Adventure of AI models
When iOS 27 is released later this year, iPhone users will reportedly have a lot of choice when it comes to which AI models they want to use on-device. Anew reportfrom Bloomberg says that Apple plans to give users their pick of a number of third-party large language models that will be capable of powering various functions within the iPhone’s operating system. The new feature, dubbed “Extensions” internally, will allow users to “access generative AI capabilities from installed apps on demand, through Apple Intelligence features such as Siri, Writing Tools, Image Playground and more,” reads a message reportedly shown in test versions of the software. The new capability will also be available for iPadOS 27 and macOS 27, the outlet writes, adding that models from Google and Anthropic are being tested now. Less clear is where ChatGPT stands. Since it is currently the model available to users, one implication is that it will remain a choice. TechCrunch reached out to Apple for more information. With long serving CEOTim Cook soon stepping down, Apple’s incoming top executive, John Ternus, is tasked with charting the future of the company — including its AI strategy. The company iswidely perceivedto be “behind” on AI, as it is not launching as many new AI services as some of its peers. However, Apple is still bringing in a generous amount of AI-based revenue and its plan is less about spending money building out AI infrastructure and services than it is about turning its already existing hardware into an AI-centric experience for users.
View
India’s first GenAI unicorn shifts to cloud services as AI model ambitions face reality
Krutrim, India’s first GenAI unicorn, is shifting from AI model development to cloud services after months of relative quiet on product updates — a move that reflects the tougher economics of building large-scale AI systems. On Tuesday, Krutrim said it was moving toward cloud services, adding that the shift follows a business overhaul in late 2025 that included reallocating capital and talent and pausing chip design efforts. The update comes more than a year after the Bengaluru-based startupreleasedits Krutrim-2 base model. The move follows a period of limited public activity from Krutrim, which has not made any significant product announcements in recent months, with itslast poston X dating back to December. The startup did not appear in any of the sessions atIndia’s AI Impact Summitin New Delhi, where global players such as Anthropic, Google, and OpenAI took part. In contrast, rival Sarvam participated in multiple sessions at the six-day AI event, where it showcasednew open source models,hardware developments, and commercial partnerships. The changes also come after a series of layoffs at Krutrim over the past year, withmore than 200 roles cutacross multiple rounds, according to local media reports. The startuppulled its Kruti AI assistant appfrom app stores in April. Founded by Bhavish Aggarwal (pictured above), who also leads ride-hailing firm Ola and EV maker Ola Electric, Krutrim had initially positioned itself as one of India’s earliest GenAI contenders, seeking to build domestic alternatives to models from companies such as Anthropic, OpenAI, and Elon Musk’s xAI. The startupraised $50 million at a $1 billion valuationin January 2024, reflecting early investor enthusiasm for India’s homegrown AI ambitions, even as AI funding in the country remains far smaller than in the U.S. Krutrim said it generated about ₹3 billion (around $31.52 million) in revenue in the financial year 2026, a threefold increase from a year earlier, along with its first annual net profit and margins exceeding 10%. The startup did not disclose how much of that revenue came from external customers versus its parent Ola’s ecosystem. Earlier reports had indicated thatabout 90% of its revenuein FY25 came from group companies. However, Krutrim said it is seeing growing demand for its AI cloud services, with more than 25 enterprise customers across sectors including telecom, financial services, and healthcare. It added that most of its GPU compute capacity is already committed to external workloads. Sanchit Vir Gogia, chief analyst at Greyhound Research, said the move toward cloud was commercially sensible, but cautioned that Krutrim’s profitability claims would need to be tested. “The standard of proof must rise with the claim,” he told TechCrunch. While Krutrim shifts toward cloud infrastructure, rivals such as Sarvam have continued to release new AI models and sign partnerships, including a recentone with space-tech firm Pixxelto develop an AI-driven orbital data center. As Gogia notes, infrastructure may be the more viable near-term play in India’s AI market, even as the longer-term ambition of building competitive models persists. Krutrim did not answer questions on its exact revenue mix, enterprise customer base, and recent restructuring.
View
4 days left: Get 50% off a second TechCrunch Disrupt 2026 pass to make more deals faster
Four days. That’s all that’s left to lock in one of the smartest advantages you can give yourself as a founder, investor, or operator right now. For the next four days only, you can buy one pass toTechCrunch Disrupt 2026andget 50% off a second of the same ticket type.That window closesMay 8 at 11:59 p.m. PT.After that, prices go up, and you’ll pay more to bring a partner or colleague.Register hereto get your plus-one pass at 50% off. Advancing from idea to IPO takes time, and how you spend that time can make a difference in whether you stall or scale. Many think it’s the pitch that slows things down. But in reality, it’s access. Fundraising is a long game of chasing proximity. Cold outreach. Missed intros. Weeks waiting for replies that never come. You spend as much time trying to get in front of the right investors as you do refining your story. Without access, capital is moving. Deals are getting done. Just not with you. WhenDisruptcomes to Moscone West in San Francisco on October 13–15, access isn’t accidental, because it’s built into the experience. Those who attend are able to access: YourDisruptticket gives you access to candid, tactical, and unfiltered insights from active founders, top-tier investors, and operators scaling real companies like: Learn more about each speaker who’ll be sharing their insights on thegrowing speaker page. And keep an eye on theDisrupt 2026 event pageto see when the agenda drops. Register nowto get 50% off a second Disrupt pass of the same type— and show up with more coverage and more opportunities to connect. WhenDisrupt hits San Francisco, more than 10,000 founders, investors, and operators, along with300+ startups, will gather with one goal: to advance deals. That changes the pace of doing business immediately. Instead of months of back-and-forth, conversations start — and move faster. You’re engaging across: You’re not burning through resources trying to get into a meeting — you’re already in one. Disrupt is apremier global startup eventwhere the ecosystem converges to move ideas, deals, and companies forward. And when youbuy a pass by May 8, you canget a second one for 50% offand accelerate that momentum even further.Register before this limited-time offer ends. AtDisrupt, you’re face-to-face with investors who can ask questions on the spot, understand how you think beyond your deck, and evaluate your vision directly. You can read signals immediately to determine what resonates, what doesn’t, and where to adjust. That kind of feedback loop compresses timelines. What normally takes weeks begins to take shape in a single day — especially as you move between sessions, meetings, and conversations across the venue. You will also find 80+ Side Events across the Bay Area for networking, workshops, and social connections, extending the value of your Disrupt ticket. Bring a second person when youbuy a second ticket for 50% offso you can multiply those moments.Get your two tickets now. Don’t settle for collecting contacts when you can come to Disrupt to connect capital to opportunity.Find your ticket matchand plan how you’ll spend your time here. With more than 20,000 curated meetings and dedicated environments like investor receptions and structured networking,Disruptlets you hear directly from founders and investors actively deploying capital and scaling companies. The value is in starting conversations that go somewhere, and bringing someone with you helps turn more of those conversations into real opportunities. Register nowto get your second ticket for 50% off. After May 8 at 11:59 p.m. PT, that price goes up — and the opportunity to bring someone takes a bigger chunk of your budget. If fundraising is already on your roadmap, waiting doesn’t make it easier. It just delays access. Buy one pass. Get 50% off the second (of the same ticket type). Bring someone who helps you move faster — and put yourself in the room where deals actually start. Secure your two passes today for Disruptand close more deals faster.
View
CopilotKit raises $27M to help devs deploy app-native AI agents
Many companies today provide AI simply as a chatbot inside their apps: You type in (or dictate) what you want it to do, and the AI bot goes and tries to do it. Still, the experience tends to feel clunky. A text-based UI doesn’t always translate to a smooth experience; for example, if you want to use a travel app to book an entire itinerary but have to scan through reams of text. According to the founders ofCopilotKit, that approach doesn’t make the most of what AI agents and LLMs can do. The company’s co-founders, Atai Barkai (pictured above, right) and Uli Barkai (pictured above, left), believe the way forward is to enable agents to live inside applications, understand what users are doing, take actions, and show useful interfaces instead of just returning long blocks of text. The company’s popularAG-UIprotocol is aimed at the first part of that solution. The widely adopted, open source protocol standardizes how AI agents connect to and communicate with user interfaces (like a web browser or an app), providing features such as streaming chat, front-end tool calls, and state sharing to enable human-in-the-loop functionality. Essentially, AG-UI gives devs the framework and tools needed to deploy AI agents within their apps. CopilotKit is also building an enterprise toolkit on top of AG-UI, adding support, self-hosted deployment features, and other must-have offerings for businesses thinking of building agents into their product. To bring that toolkit to market, the Seattle-based startup has raised $27 million in a Series A round led by Glilot Capital, NFX, and SignalFire, TechCrunch has exclusively learned. The flexible user interface is a particular selling point. CEO Atai Barkai told TechCrunch developers can use the startup’s framework to provide the specifications and building blocks for dynamic user interfaces, which an AI agent can then use to generate UIs to fit the context. “The agent can reply to you, not just with blocks of text, but with interactive UIs that are defined by your own company,” Atai explained. “If, for example, a user asks for breakdown of revenue by category, instead of getting this kind of big, impenetrable paragraph, you get a pie chart, and it’s your own design of the pie chart that the user can interact with […] So all of your agents can, very trivially, speak to a UI and use these catalog of components and show that to users.” Atai also noted that CopilotKit’s toolkit gives developers full control over how much their AI agent can change the UI, to the point where they can choose to have the interface be “pixel-perfect” or just provide broad building blocks that the AI can put together as required. The funding follows a period of strong adoption both for AG-UI and CopilotKit. The protocol, which works alongside the widely adoptedModel Context Protocol (MCP)andAgent2Agent (A2A) protocol, is today supported by major AI infrastructure providers likeGoogle,Microsoft,Amazon, andOracle, as well as popular frameworks likeLangChain,Mastra,PydanticAI, andAgno. Atai said CopilotKit and AG-UI (the company’s strongest claim to ecosystem relevance) see millions of installs per week, and that a large portion of Fortune 500 companies are using the protocol and the startup’s tools in production. Meanwhile, CopilotKit counts enterprise bigwigs like Deutsche Telekom, Docusign, Cisco, and S&P Global as enterprise customers. To tap that growing interest, the company is also launching CopilotKit Enterprise Intelligence, a self-hostable offering that bundles a number of infrastructure features to fully deploy agents within apps. CopilotKit faces heated competition in the market for enterprise agents tools. Cloud platform Vercel’s open sourceAI SDKhelps developers build AI applications with similar capabilities, andassistant-uioffers components for building AI chat interfaces. Meanwhile, OpenAI’s Apps SDK is also an option for building richer interfaces, though only inside ChatGPT. Atai argues that CopilotKit is different from those offerings because it takes a horizontal, enterprise-friendly approach rather than a vertically integrated one. Instead of offering a full-stack AI platform, CopilotKit aims to support whatever agent framework, cloud provider, or back end an enterprise already uses. “If there are two things we hear in almost every single enterprise conversation, enterprises want optionality and they want self-hosting,” he said. “Maybe they’re already using the Google, Amazon, Oracle, Microsoft, LangChain, Mastra stacks. They want optionality, and they want self-hosting, and these are two things that they don’t really get in the Vercel stack.” That open positioning will be important to maintain. Companies that build on top of their own open source infrastructure often face a tension, which is that they want their technology to stay a neutral standard, but they also need to build a business on top of it. But Atai said that AG-UI is a fully open protocol, and that CopilotKit’s commercial product is meant to harden the open source stack for enterprises, not replace it. “They’re very much complementary. Our strategy is to be the default choice in the ecosystem, and then to monetize the top enterprises,” Uli, the startup’s head of growth, added. “So it’s very much in our interest that the open source is the best out there, and the 95% of users can just go build and get started without paying anyone or talking to anyone.” The company currently has about 25 employees and plans to use the new funding to grow its team.
View
ElevenLabs lists BlackRock, Jamie Foxx and Eva Longoria as new investors
Voice AI company ElevenLabsrevealednew investors that are part of its $500 million Series D fundraise,which was first announced in February. The additions include institutions such as BlackRock, Wellington, D.E. Shaw, and Schroders; enterprises like NVIDIA, Salesforce, Santander, KPN, and Deutsche Telekom; and individual investors such as Jamie Foxx, Eva Longoria, and Squid Game creator Hwang Dong-hyuk. The startup also noted that it surpassed $500 million in ARR (Annual Recurring Revenue), after endinglast year with nearly $350 million in ARR. The company’s co-founder and CEO, Mati Staniszewski, said last month that ElevenLabs added$100 million in net new ARR in Q1 2026, ending the quarter at roughly $450 million in ARR. The company has also accelerated its valuation rapidly, growing from $6.6 billion last September to $11 billion this February. “Voice is the highest-stakes channel for any customer interaction, and the bar for quality, latency, and security is extremely high. ElevenLabs is not just a category leader – it is becoming a foundational enabler of Deutsche Telekom’s broader Industrial AI vision. From voice-as-a-service to multilingual automation and in-network AI agents, we believe the company is uniquely positioned to reshape how businesses interact with customers across all channels,” Karine Peters, Managing Director at Deutsche Telekom’s venture arm T. Capital said in a statement. In the past quarter, the voice AI company has signed enterprise contracts with the likes of Deutsche Telekom, Revolut, and Klarna. ElevenLabs said that, besides the fundraising, it also closed a $100 million tender, a second in roughly six months after the company issued one last September. Staniszewskisaid in a blog postthat the company will give an opportunity to retail investors to invest in ElevenLabs through Robinhood Ventures, but didn’t provide details about the program. Staniszewski noted that consumers won’t trust systems that sound robotic or “interact strangely” and emphasized the importance of building “human-level AI voice models.” Last month, the company acquired the team from Polish voice AI startup Paplato bolster its research team.
View
Meta will use AI to analyze height and bone structure to identify if users are underage
Meta will start using AI to scan photos and videos for visual clues to see if a user is under 13 and should be removed from Facebook and Instagram, the companyannouncedon Tuesday. These visual clues include a person’s height or bone structure, it said. “We want to be clear: this is not facial recognition,” Meta explained in its blog post. “Our AI looks at general themes and visual cues, for example height or bone structure, to estimate someone’s general age; it does not identify the specific person in the image. By combining these visual insights with our analysis of text and interactions, we can significantly increase the number of underage accounts we identify and remove. The visual analysis system is now operating in select countries, but Meta says it’s working toward a broader rollout. Meta says this system is part of its efforts to keep kids under 13 off its platforms. These efforts include using AI to analyze entire profiles for contextual clues, such as birthday celebrations or mentions of school grades. The company looks for these signals across different formats, such as posts, comments, bios, captions, and more. Meta plans to expand this technology to more parts of its apps, including Instagram Live and Facebook Groups, in the future. If Meta determines that a person may be underage, it will deactivate their account, and the user will need to prove their age using the company’s age verification process in order to prevent their account from being deleted. The announcement comes weeks after a New Mexico juryordered Meta to pay $375 millionin civil penalties for misleading consumers about the safety of its platforms and putting children at risk. The company was also ordered to implement fundamental changes to its platforms. Meta has since threatened toshut down its social media servicesin the state. It’s worth noting that this case isone of manylawsuits that Meta and other Big Tech companies are facing over child safety. Meta also announced on Tuesday that it’s expanding its technology that automatically places teens into stricter“Teen Accounts”on Instagram to 27 countries in the EU and Brazil. These teen accounts place users into a stricter account experience with additional safeguards, such as receiving DMs only from people they follow or are already connected to, hiding harmful comments, and setting accounts to private by default. Additionally, Meta said it’s expanding the technology to Facebook in the U.S. for the first time, followed by the U.K. and EU in June.
View
Etsy launches its app within ChatGPT as it continues its AI push
EtsyannouncedTuesday the launch of its native app within ChatGPT, opening up a new way for shoppers to explore its catalog of over 100 million listings. The new experience is designed to move beyond the limitations of traditional keyword queries. Instead of typing something like “wooden coffee table,” then scrolling and adjusting filters, users can now express what they’re looking for in natural language. For instance, “Help me find a Mother’s Day gift under $100 for my mom who loves gardening.” Now live in beta, the feature allows users to tag @Etsy directly within a prompt. From there, the Etsy app in ChatGPT surfaces relevant product listings that users can browse, compare, and click through to Etsy for additional details or purchase. This isn’t Etsy’s first experiment inside ChatGPT. Back in September, Etsy became an early partner in ChatGPT’sInstant Checkoutintegration, which let users buy products directly inside the chat interface. However, the initiativeended in March, suggesting it didn’t perform as OpenAI had hoped. It wasreportedthat Etsy didn’t see a large volume of sales from the integration, leading Etsy to start building a native app within ChatGPT instead. Alongside this launch, Etsy also revealed it’s testing a beta conversational search experience within its platform, specifically geared toward helping users find gifts. The gift assistant acts as a personal shopper, offering a guided, conversational way to discover ideas, narrow down preferences, and surface relevant products. This builds on Etsy’s broader AI push, which includes anAI-powered discovery experiencefeaturing curated collections and a suite ofseller tools, including a tool that helps generate product titles and descriptions, as well as a writing assistant to help draft messages to buyers. In 2024, Etsyintroduceda new “Designed” label to identify AI content, part of an effort to increase transparency as AI-generated artwork becomes more prevalent on the platform. The news of a ChatGPT integration comes a week after Etsyreportedits Q1 2026 earnings, surpassing revenue expectations with $631 million, and marketplace gross merchandise sales were up 6% year over year. Notably, active buyers increased for the first time in two years to 86.6 million. Etsy also touted 5.6 million active sellers on the platform. In February, the companyannouncedit was selling Depop to eBay for $1.2 billion in cash, a move aimed at doubling down on its core marketplace. Etsy joins a growing list of companies buildingnative appswithin ChatGPT, including Angi, SeatGeek,Tubi, and Wix. Developers have been able to build apps within the chatbot since October.
View
PayPal says it’s ‘becoming a technology company again.’ That means AI.
PayPal is looking towards the future, despite its falling stock and looming layoffs. In its first-quarter earnings call, CEO Enrique Lores told investors that PayPal needs to “recommit to the fundamentals,” which included “becoming a technology company again.” There was no need to read between the lines — PayPal was pitching an AI-powered turnaround. Lores explicitly said so, telling analysts on this week’s call that leading companies find ways to differentiate themselves by innovating, and that now is the time for PayPal to take action. This includes modernizing its tech platform, moving faster to become “cloud-native,” and “aggressively adopting AI in our development processes,” Lores said. The latter would increase developer productivity and shorten time to market, he added. It’s a startling admission from PayPal that it has yet to fully embrace AI in-house, when AI-assisted coding is one of the breakout areas where the technology has truly excelled. Other consumer tech companies have rapidly adopted AI in recent months to assist with coding, with Spotify even declaring in February that its top developershaven’t written a line of codesince December. Meanwhile, top dev teams are trying to outcompete one anotherby tokenmaxxing— a proxy for understanding who at the company is experimenting with AI more often, based on the number of AI tokens they use. PayPal is only now catching up, it seems. Lores said the company has formed a new “AI transformation and simplification” team to help with its enterprise AI agenda. Combined with the planned layoffs, which Lores characterized as PayPal removing layers from its organizational structure, the addition of AI-enabled processes is expected to bring the company at least $1.5 billion in cost savings over the next two to three years, he said. The company announced last week it wasreorganizing its business, which streamlines the operation into three segments: checkout solutions and PayPal, consumer financial services (and Venmo) as well as payment services and crypto. In addition,Bloomberg reportedon Tuesday that PayPal plans to cut around 20% of its workforce over the next two to three years as part of its cost-savings plan, equating to north of 4,500 jobs. More cost savings will come from PayPal’s plans for AI adoption, company execs said on the call. That includes bringing AI into areas beyond coding, like customer service, support operations, and risk management, to name a few. “I think the changes that AI will enable us to do are going to drive — are going to be very significant,” said Lores. “This is why we created a group last week, reporting to me, that is going to be in charge of driving — function by function, process by process — this AI transformation. And this is not about adopting AI as a technology, where we have done many pilots in the company, and we have seen what is possible. It’s really about understanding how can we redesign the key processes … this is what we have seen that really will drive significant savings.” Announcing an AI-driven push to cut costs while eliminating thousands of jobs underscores a core criticism of the technology — it comes with a human cost It’s worth noting that, in this case, PayPal was already in need of restructuring. The company may have beat on its first-quarter earnings with revenue of $8.4 billion, up 7% year-over-year, but it forecast weak guidance for the second quarter,sending the stock tumblingafter earnings. That follows a long post-pandemic decline that has sentthe stock down over 80%from its 2021 high and has stunted PayPal’s growth. Asked if separating Venmo into its own business meant the company would be open to selling it, Lores said that, for now, this is what made the most sense in terms of the turnaround plan. Still, he signaled openness to future deals by saying “my number one priority is to maximize shareholder value,” in answer to an analyst’s question about a sale.
View
