CodeCanyon

Applications are almost closed, and you have until 11:59 p.m. PT tonight to reserve your exhibitor table at TechCrunch Sessions: AI, our premiere industry event that’s happening at UC Berkeley’s Zellerbach Hall on June 5. Imagine walking into a room filled with 1,200+ investors, founders, enterprise leaders, and journalists — all hunting for the next big thing. Will they discover you, or your competitor? Exhibiting is more than a table — it’s a high-leverage move. Consider the perks: Getting discovered by the exact people you want to meet. Positioning your brand as a serious player in AI. Sparking conversations that lead to investment, partnerships, and press. What comes with your exhibit: Prime 6′ x 3′ table with signage and full setup. Branded profile in the official event app. 5 full-access tickets for your team. Lead capture tools to collect contacts. Wi-Fi, power access, and marketing visibility. This is the only AI-focused TechCrunch event of the year — and it’s your best bet to get right in front of the industry’s movers and shakers in a more intimate setting than the numerous other crammed AI events. Remember, that final deadline is tonight, May 9 at 11:59 p.m. PT. You’ve come too far to go unnoticed. Make your move now and let your innovation lead the future. Techcrunch event Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 BOOK NOW Book your table now and take your place in the spotlight.

Josh Raffaelli, who has long roots as a Silicon Valley investor and has backed a number of Elon Musk companies, is suing his former employer, the massive trillion-dollar AUM Brookfield Asset Management, reports the New York Times. Much of Raffaelli’s complaint concerns how Brookfield covered pandemic-related real estate losses and alleges the company fired him after he filed a whistleblower complaint at the SEC. His suit makes allegations like fraud and bribery, while Brookfield vehemently denies any wrongdoing, it told the Times. In February, Brookfield quietly shuttered the venture capital unit run by Raffaelli and rolled some assets into another unit, Bloomberg reported at the time. One of Raffaelli’s complaints in the suit is that Brookfield didn’t buy as much stock in Musk-owned companies as he had secured the ability to buy. Raffaelli had deals to buy into Musk companies like SpaceX, xAI and the Boring Company, the suit alleges. Raffaelli’s Brookfield fund was, however, a big backer of Musk’s takeover of Twitter, Bloomberg reported. The lawsuit is a very public battle for Raffaelli, who previously worked as a partner at the VC firm then known as Draper Fisher Jurvetson. (Today, it’s a collection of funds.) While at DFJ, Brookfield helped that firm’s make investments into Musk companies like SolarCity (acquired by Tesla), SpaceX, and Tesla.

Microsoft employees aren’t allowed to use DeepSeek due to data security and propaganda concerns, Microsoft vice chairman and president Brad Smith said in a Senate hearing today. “At Microsoft we don’t allow our employees to use the DeepSeek app,” Smith said, referring to DeepSeek’s application service (which is available on both desktop and mobile.) Smith said Microsoft hasn’t put DeepSeek in its app store over those concerns, either. Although lots of organizations and even countries have imposed restrictions on DeepSeek, this is the first time Microsoft has gone public about such a ban. Smith said the restriction stems from the risk that data will be stored in China and that DeepSeek’s answers could be influenced by “Chinese propaganda.” DeepSeek’s privacy policy states it stores user data on Chinese servers. Such data is subject to Chinese law, which mandates cooperation with the country’s intelligence agencies. DeepSeek also heavily censors topics considered sensitive by the Chinese government. Despite Smith’s critical comments about DeepSeek, Microsoft offered up DeepSeek’s R1 model on its Azure cloud service shortly after it went viral earlier this year. Techcrunch event Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 BOOK NOW But that’s a bit different from offering DeepSeek’s chatbot app itself. Since DeepSeek is open source, anybody can download the model, store it on their own servers and offer it to their clients without sending the data back to China. That, however, doesn’t remove other risks like the model spreading propaganda or generating insecure code. During the Senate hearing, Smith said that Microsoft had managed to go inside DeepSeek’s AI model and “change” it to remove “harmful side effects.” Microsoft did not elaborate on exactly what it did to DeepSeek’s model, referring TechCrunch to Smith’s remarks. In its initial launch of DeepSeek on Azure, Microsoft wrote that DeepSeek underwent “rigorous red teaming and safety evaluations” before it was put on Azure. While we can’t help pointing out that DeepSeek’s app is also a direct competitor to Microsoft’s own Copilot internet search chat app, Microsoft doesn’t ban all such chat competitors from its Windows app store. Perplexity is available in the Windows app store, for instance. Although any apps by Microsoft’s archrival Google (including the Chrome browser and Google’s chatbot Gemini) did not surface in our webstore search.

Autonomous vehicle technology company Aurora Innovation plans to expand on the success of its first driverless commercial launch and add night driving to its operations. Aurora said Thursday in the second half of 2025 it will start sending its self-driving trucks out at night and during adverse weather conditions like rain or heavy wind. The company, which provided the update in its first-quarter shareholder letter, also plans to expand its driverless trucking route beyond Dallas to Houston, and into El Paso and Phoenix. Aurora already runs freight with self-driving trucks in those conditions, but with a human safety operator behind the wheel. The company said it has completed more than 4,000 miles in a single self-driving truck without a driver ruuning freight for its launch customers Hirschbach Motor Lines and Uber Freight. In the week since Aurora’s commercial launch, the company has already expanded to two driverless trucks operating on a daily basis, and says it expects to operate “tens of trucks” by the end of 2025. The milestone, and future plans come alongside another major shift at the company: the resignation of co-founder and chief product officer Sterling Anderson. Aurora shared new details Thursday in its first-quarter shareholder letter about plans to grow its autonomous freight service, signaling it will offer more specific timelines for key milestones as it expands. Aurora reported $211 million in operating expenses, including $153 million for R&D. It used $142 million in operating cash and $8 million in capex in the first quarter, ending with nearly $1.2 billion in cash and short-term investments. Aurora expects to spend $175 to $185 million per quarter for the rest of this year. The company didn’t provide revenue figures in its Q1 report. Techcrunch event Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 BOOK NOW In the short-term, Aurora plans to own, operate, maintain and insure its own trucks – made available on the Uber Freight network – for customers. The company is working with partners Paccar and Volvo Trucks to build self-driving trucks at scale. Starting in 2027 or earlier, Aurora expects customers to buy those trucks directly from manufacturers so it can shift to a driver-as-a-service model.

In April, South Korea’s telco giant SK Telecom (SKT) was hit by a cyberattack that led to the theft of personal data on approximately 23 million customers, equivalent to almost half of the country’s 52 million residents. At a National Assembly hearing in Seoul on Thursday, SKT chief executive Young-sang Ryu said about 250,000 users have switched to a different telecom provider following the data breach. He said that expects this number to reach 2.5 million, more than tenfold the current amount, if the company waives cancellation fees. The company could lose up to $5 billion (around ₩7 trillion) over the next three years if it decides not to charge cancellation fees for users who want to cancel their contract early, Ryu said at the hearing. “SK Telecom considers this incident the most severe security breach in the company’s history and is putting forth our utmost effort to minimize any damage to our customers,” a spokesperson at SKT told TechCrunch in an emailed statement. “The number of customers affected and the entity responsible for the hacking is under investigation,” the spokesperson added. A joint investigation involving both public and private entities is currently underway to identify the specific cause of the incident. The Personal Information Protection Committee (PIPC) of South Korea announced on Thursday that 25 different types of personal information, including mobile phone numbers and unique identifiers (IMSI numbers), as well as USIM authentication keys and other USIM data, had been exfiltrated from its central database, known as its home subscriber server. The compromised data can put customers at greater risk of SIM swapping attacks and government surveillance. After its official announcement of the incident on April 22, SKT has been offering SIM card protection and free SIM card replacements to prevent further damage to its customers. “We detected possible information leakage regarding SIM on April 19,” the spokesperson at SKT told TechCrunch. “Following the identification of the breach, we immediately isolated the affected device while thoroughly investigating the entire system.” “To further safeguard our customers, we are currently developing a system that can protect users’ information through the SIM protection service while allowing them to use roaming services seamlessly outside of Korea by May 14,” the spokesperson said. To date, SKT has not received any reports of secondary damage and no verified instances of customer information being distributed or misused on the dark web or other platforms, the company told TechCrunch. A timeline of SKT’s data breach April 18, 2025 SKT detected abnormal activities on April 18 at 11:20 pm local time. SKT found unusual logs and signs of files having been deleted on equipment that the company uses for monitoring and managing billing information for its customers, including data usage and call durations. April 19, 2025 The company identified a data breach on April 19 in its home subscriber server in Seoul, which typically houses subscriber information, including authentication, authorization, location, and mobility details. April 20, 2025 SKT reported the cyberattack incident to Korea’s cybersecurity agency on April 20. April 22, 2025 SKT confirmed on its website that it detected suspicious activity, indicating a “potential” data breach involving some information related to users’ USIMs data. April 28, 2025 SKT began replacing mobile SIM cards of 23 million users, but the company has faced shortages in obtaining sufficient USIM cards to fulfill its promise to provide free SIM card replacements. April 30, 2025 South Korean police began investigating SKT’s suspected cyberattack on April 18. April 30, 2025 South Korean police began investigating SKT’s cyberattack on April 30. According to local media reports, many South Korean companies, including SKT, use Ivanti VPN equipment, and that the recent data breach may be connected to China-backed hackers. Per a local media report, SKT said it received a cybersecurity notice from KISA instructing the company to turn off and replace the Ivanti VPN. TeamT5, a cybersecurity company based in Taiwan, alerted the public to the worldwide threats posed by a government-backed group linked to China, which allegedly took advantage of vulnerabilities in Ivanti’s Connect Secure VPN systems to gain access to multiple organizations globally. Some 20 industries have been affected, including automotive, chemical, financial institutions, law firms, media, research institutes, and telecommunications, across 12 countries, including Australia, South Korea, Taiwan, and the United States. May 6, 2025 A team of public and private investigators discovered an additional eight types of malware in SKT’s hacking case. The team is currently investigating whether the new malware was installed on the same home subscriber server as the original four strains or if they are located on separate server equipment. May 7, 2025 Tae-won Chey, the chairman of SK Group, which operates SKT, publicly apologized for the first time for the data breach, some three weeks after the breach occurred. As of May 7, all eligible users have been signed up for the SIM protection service, except those living abroad using roaming services and temporarily suspended, the spokesperson told TechCrunch, adding that its fraud detection system has already been set up for all customers to prevent unauthorized login attempts using cloned SIM cards. May 8, 2028 SKT is currently assessing how to handle the cancellation fees for users affected by the data breach incident. About 250,000 users have switched to another telecom provider following the breach, according to the company’s chief executive at a National Assembly hearing. South Korean authorities, meanwhile, announced that 25 types of personal information were leaked from the company’s databases during the cyberattack.

Sterling Anderson, a veteran of the nascent autonomous vehicle sector and co-founder of Aurora, is resigning just a week after the company launched its commercial self-driving truck service in Texas. Anderson held the chief product officer position at Aurora. The resignation was posted in a regulatory filing along with the company’s first-quarter earnings report. His resignation will go into effect June 1. He will leave the board August 31. The company said in the filing that his resignation from the board “did not result from any disagreement with the Company concerning any matter relating to its operations, policies, or practices. The Company and the entire Board are deeply grateful for Mr. Anderson’s service and his immense contributions to the Company over the years in his role as founder, Chief Product Officer and a member of the Board.” Anderson could not be reached for comment. Anderson was director of Tesla’s Autopilot program when he left co-found Aurora in 2017 alongside CEO Chris Urmson, the former head of the Google self-driving project, and Drew Bagnell, who was leading Uber’s autonomy and perception team. The trio, considered pioneers of the autonomous vehicle industry, gave Aurora immediate buzz, helping it attract high-profile investors like Sequoia Capital, Amazon, and T. Rowe Price Associates as well as a slew of partnerships. It gained more cache in December 2020 when it reached an agreement with Uber to buy the ride-hailing firm’s self-driving unit in a complex deal that valued the combined company at $10 billion. Under the terms of that acquisition, Aurora did not pay cash for Uber ATG, a company that was valued at $7.25 billion following a $1 billion investment in 2019 from Toyota, DENSO and SoftBank’s Vision Fund. Instead, Uber handed over its equity in ATG and invested $400 million into Aurora. Uber received a 26% stake in the combined company, according to a filing with the U.S. Securities and Exchange Commission. Within four years the company went from buzzy startup to publicly traded company via a merger with special purpose acquisition company Reinvent Technology Partners Y. The SPAC was launched by LinkedIn co-founder and investor Reid Hoffman, Zynga founder Mark Pincus and managing partner Michael Thompson. This story is developing… Techcrunch event Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 BOOK NOW

Over $10 billion — that’s how much revenue Apple’s U.S. App Store raked in last year, according to a new analysis by app intelligence provider Appfigures. The firm’s estimates indicate that U.S. App Store revenue from commissions more than doubled between 2020 and 2024. In 2020, Apple’s share of App Store commissions was approximately $4.76 billion, growing to over $10.1 billion by 2024. Based on Appfigures’ data, U.S. App Store developers generated $33.68 billion in gross revenue from their apps and games using Apple’s payments system in 2024, and took home $23.57 billion after Apple’s cut. Image Credits:AppfiguresThough Apple doesn’t typically break out its App Store revenue during earnings, it did publish a report in May 2023 where it said the App Store globally generated $104 billion in estimated billings for digital goods and services in 2022. However, Appfigures’ analysis found the App Store made $61.5 billion globally in 2022, which grew to $91.3 billion in 2024. From this, Apple made more than $27.39 billion in commissions globally last year, Appfigures also said. That leads to a discrepancy between Appfigures’ analysis and Apple’s own. This can be explained by an important caveat found in Apple’s report. Under Apple’s chart, it states that its “billings and sales” figures are “not the same as App Store billings.” That’s important here. Techcrunch event Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 BOOK NOW When Apple wrote its report, the company was trying to show how big the App Store is and how key it is to the overall economy, so it merged App Store revenue with revenue generated outside the App Store to generate its total for the “Billings and Sales” category. In the report, Apple calculated the portion of an app’s total revenue that is facilitated by the App Store, even if the purchase was made elsewhere. For instance, if a user buys a subscription to Hulu on the web, but then spends 60% of their time streaming Hulu on Apple devices, Apple credits itself with facilitating 60% of that user’s spend. (To determine usage, the report relied on third-party sources, like market research firms, to estimate how much usage occurred on smartphones versus tablets, desktops, or TVs.) Apple also allows enterprises to distribute apps with in-app purchases, but these aren’t visible in the App Store. “Grave Irreparable Harm?” Examining the numbers around U.S. Apple App Store revenue is more relevant than ever in the wake of the recent court ruling that now prevents Apple from charging a 27% commission on transactions that take place outside the App Store. Apple initially attempted to comply with the court’s injunction resulting from its antitrust battle with Fortnite maker Epic Games by making changes that wouldn’t harm App Store profits. To do so, Apple last year gave developers a way to apply for an exception to its App Store rules so they could add web links inside their apps that directed customers to external purchases. However, Apple continued to charge a 27% commission on those purchases and dictated how the website links should appear. (This even included the use of “scare screens” to warn consumers of the dangers of making purchases outside the App Store.) Last week, a judge ruled that Apple was in “willful violation” of the 2021 injunction by continuing to collect fees on purchases made outside apps and by creating new anticompetitive barriers. This decision forced Apple to update its U.S. App Store rules, which now allow developers to link out to other ways for consumers to make purchases, without obstacles or commissions. Since then, several apps have taken advantage of the ability to introduce web payments, including Spotify, Amazon Kindle, and Patreon. One small game emulator called Delta is now supporting itself via Patreon memberships, too. Apple is appealing the decision, arguing in its most recent filing that the ruling causes Apple “grave irreparable harm.” “These restrictions, which will cost Apple substantial sums annually, are based on conduct that has never been adjudicated to be (and is not) unlawful,” Apple’s filing stated. “Rather, they were imposed to punish Apple for purported non-compliance with an earlier state-law injunction that is itself invalid.” This argument won’t likely go over well with developers, as many believe Apple should have lowered commissions for everyone years ago, not just for small business developers. Appfigures’ analysis also broke down U.S. App Store revenue by apps and games, which generated Apple approximately $6.28 billion and $3.83 billion, respectively, in 2024. Together, these figures highlight how critical App Store revenue remains to Apple’s bottom line, and why it’s fighting so hard to retain control. Apple tries to delay ruling that bars it from taking a cut on external app payments Apple wins antitrust court battle with Epic Games, appeals court rules

OpenAI is enhancing its AI-powered “deep research” feature with the ability to analyze codebases on GitHub. On Thursday, OpenAI announced what it’s calling the first “connector” for ChatGPT deep research, the company’s tool that searches across the web and other sources to compile thorough research reports on a topic. Now, ChatGPT deep research can link to GitHub (in beta), allowing developers to ask questions about a codebase and engineering documents. The connector will be available for ChatGPT Plus, Pro, and Team users over the next few days, with Enterprise and Edu support coming soon, according to an OpenAI spokesperson. OpenAI’s ChatGPT Deep Research feature can now connect to GitHub.Image Credits:OpenAIThe GitHub connector for ChatGPT deep research arrives as AI companies look to make their AI-powered chatbots more useful by building ways to link them to outside platforms and services. Anthropic, for example, recently debuted Integrations, which gives apps a pipeline into its AI chatbot Claude. OpenAI years ago offered a plugin capability for ChatGPT, but deprecated it in favor of custom chatbots called GPTs. “I often hear that users find ChatGPT’s deep research agent so valuable that they want it to connect to their internal sources, in addition to the web,” OpenAI Head of Business Products Nate Gonzalez wrote in a blog post on LinkedIn. “[That’s why] today we’re introducing our first connector.” In addition to answering questions about codebases, the new ChatGPT deep research GitHub connector lets ChatGPT users break down product specs into technical tasks and dependencies, summarize code structure and patterns, and understand how to implement new APIs using real code examples. Techcrunch event Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 BOOK NOW There’s a risk that ChatGPT deep research hallucinates, of course — no AI model in existence doesn’t confidently make things up sometimes. But OpenAI is pitching the new capability as a potential time saver, not a replacement for experts. An OpenAI spokesperson said ChatGPT will respect an organization’s settings so users only see GitHub content they’re already allowed to view and codebases that’ve been explicitly shared with ChatGPT. OpenAI has been investing in its tooling for assistive coding, recently unveiling an open source coding tool for terminals called Codex CLI and upgrading the ChatGPT desktop app to read code in a handful of developer-focused coding apps. The company sees programming as a top use case for its models. Case in point, OpenAI has reportedly reached an agreement to buy AI-powered coding assistant Windsurf for $3 billion. In other OpenAI news on Thursday, the company launched fine-tuning options for developers looking to customize its newer models for particular applications. Devs can now fine-tune OpenAI’s o4-mini “reasoning” model via a technique OpenAI calls reinforcement fine-tuning, which uses task-specific grading to improve the model’s performance. Fine-tuning has also rolled out for the company’s GPT-4.1 nano model. Only verified organizations can fine-tune o4-mini, according to OpenAI. GPT-4.1 nano fine-tuning, meanwhile, is available for all paying developers. OpenAI began gating certain models and developer features behind verification, which requires organizations to submit an ID and other identity documents, in April. The company claims that it’s necessary to prevent abuse.

Social media startup Fizz is suing grocery delivery giant Instacart and party planning app Partiful for trademark infringement, the company announced on Thursday. Earlier this week, Instacart launched a new drinks and snack delivery app for parties called Fizz and announced that Partiful had integrated Fizz directly into its platform. Founded in 2020, Fizz is a Gen Z-focused social networking app available on more than 400 college campuses. The suit, filed Wednesday in the U.S. District Court for the Northern District of California, seeks a jury trial, injunctive relief, damages, and a court order barring Instacart and Partiful from using the “FIZZ” name in connection with social or event planning services. Instacart and Partiful did not respond to TechCrunch’s request for comment. In the lawsuit, Fizz states that it has been using the “FIZZ” trademark since January 2022 and filed for trademark registration in December 2021. The startup is accusing Instacart and Partiful of common law trademark infringement, federal trademark infringement, cybersquatting, and violating California’s unfair competition laws. Image Credits:Instacart“This new Fizz App by Instacart and Partiful is a blatant attempt to misappropriate the goodwill that Plaintiff has painstakingly developed through its continuous use of the FIZZ Marks among the Gen-Z demographic,” the lawsuit reads. “Together, Instacart and Partiful are competing head-on with Plaintiff in its core market of event planning for the Gen-Z demographic. Instacart and Partiful could have chosen any name for their new venture, but rather than compete on an even playing field, they are using FIZZ.” Fizz alleges that Instacart and Partiful knowingly launched the new app with the identical name for the same Gen Z demographic, creating a likelihood of confusion among customers who may believe that the new ordering service is affiliated with or endorsed by Fizz. Techcrunch event Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 BOOK NOW The startup also alleges that Instacart and Partiful are exploiting its brand recognition as a known social platform for Gen Z. “Plaintiff is informed and believes and thereupon alleges that Defendant Instacart had a bad faith intent to profit from the FIZZ Marks when it registered the domain name ,” the lawsuit states. “Specifically, Defendant Instacart knew or should have known of the FIZZ Marks and incorporated Plaintiff’s trademark and trade name in its domain name. In doing so, Defendant Instacart intended to divert consumers from the Fizz Platform’s online location to the Fizz App online location for Defendant’s own commercial gain.” Additionally, the lawsuit alleges that Partiful competes with Fizz directly in the events planning space, and that the company is now using Fizz’s name to confuse the Gen Z demographic after failing “to win the Gen Z market through fair competition.” The lawsuit announced today isn’t Fizz’s first brush with legal action, as the startup sued rival Sidechat in 2023 over unfair competition practices.

Sankaet Pathak’s last startup, fintech Synapse, filed for bankruptcy in 2024 amid issues with partner Evolve Bank & Trust. Tens of millions of dollars in deposits made by consumers, mostly customers of fintechs that worked with Synapse, remain unaccounted for. Yet according to The Information, Pathak is reportedly moving full steam ahead on attempts to fundraise for his new venture, humanoid robotics startup Foundation. Pathak is said to be in the midst of raising $100 million for Foundation at a whopping $1 billion valuation. The numbers seem particularly ambitious considering the startup only debuted its humanoid robot, Phantom, earlier this year. Foundation only last August raised $11 million in a pre-seed funding round from Tribe Capital and “other angels.” Foundation’s self-proclaimed mission is to “create advanced humanoid robots that can operate in complex environments” to address the labor shortage. TechCrunch has reached out to Pathak for comment.

Meta has chosen Robert Fergus to lead its Fundamental AI Research (FAIR) lab, according to Bloomberg. Fergus had been working at Google DeepMind as a research director for nearly five years, per his LinkedIn. Prior to Google, he worked as a researcher scientist at Meta. Meta’s FAIR, which has been around since 2013, has faced challenges in recent years, according to a report from Fortune. FAIR led research on the company’s early AI models, including Llama 1 and Llama 2. However, researchers have reportedly departed the unit en masse for other startups, companies, and even Meta’s newer GenAI group, which led the development of Llama 4. Meta’s previous VP of AI Research, Joelle Pineau, announced in April she’d be leaving the company for a new opportunity.

Instagram Threads will begin testing video ads, Meta announced on Thursday. The test, which will make Threads look more like its competitor X, is an expansion of Threads’ advertising initiatives, which began last month with the opening up of ads to global advertisers. The news was announced at Meta’s presentation at the IAB NewFronts, where a number of social media companies pitch themselves to advertisers. On Threads, Meta says a “small number” of advertisers will test 19:9 or 1:1 video ad creatives that will appear in between pieces of organic content in the Threads feed. The company didn’t share other details around pricing or frequency of those ads, however. Image Credits:Meta's Instagram ThreadsThe update follows Meta’s recent announcement that Threads now reaches over 350 million monthly active users. The app has also seen a 35% increase in the time spent on Threads as a result of improvements to the app’s recommendation systems, Meta CEO Mark Zuckerberg also told investors on Meta’s earnings call in April. Meta announced the news around Threads, among other updates to its ad products, at the NewFronts. The company says it’s also testing a new short-form video solution, Reels trending ads, that will be shown next to the most trending Reels from creators. Rival TikTok this week had also introduced an expansion of its similar offering, called Pulse Suite, which will now let advertisers market themselves next to trending content by category, holiday, tentpole moments, cultural events, and evergreen, always-on content from sports, entertainment, and lifestyle publishers. Techcrunch event Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 BOOK NOW Meta will also begin to test Trends in Instagram’s Creator Marketplace, to help advertisers find popular trneds, and will test the Creator Marketplace API to help businesses find and connect with quality creators at scale. The company is also rolling out Video Expansion on Facebook Reels, which adjusts video assets by generating unseen pixels in each video frame to expand the aspect ratio for a more native experience.

Months after the hacked education software maker PowerSchool paid a hacker’s ransom to delete the company’s banks of stolen student data, at least one school district says it is now being extorted by someone who said the data was not destroyed. PowerSchool, which provides its K-12 software to thousands of schools to support 60 million students across North America, was hacked in December 2024 using a single stolen credential, which allowed a hacker broad access to PowerSchool’s stores of personally identifiable student and teacher data, including Social Security numbers and health data. The company said at the time that it had paid the hacker a ransom to allegedly delete the stolen data, but it has repeatedly refused to disclose the sum it paid. Now, Toronto’s district school board, which serves around 240,000 students each year, said in a statement that earlier this week it had “received a communication from a threat actor demanding a ransom using data from the previously reported incident.” Several other schools in North America received extortion notes, including across North Carolina, per local media. PowerSchool confirmed that it had paid the ransom at the time, saying the company “thought it was the best option for preventing the data from being made public.” Some cybersecurity professionals and law enforcement have long discouraged victims from paying a ransom as there are no guarantees that the hackers will stick to their word when claiming to delete stolen data. As evidenced by past ransomware and extortion incidents, some gangs were later found to have retained huge amounts of stolen victim data, often to revictimize affected individuals with additional extortion attempts. In a statement shared with customers this week, seen by TechCrunch, PowerSchool said it “recently became aware that a threat actor has reached out to some PowerSchool SIS customers in an attempt to extort them using data” from the December 2024 breach. Beth Keebler, a spokesperson for PowerSchool, told TechCrunch that the company does not think this is a new incident because “samples of data match the data previously stolen in December.” PowerSchool has not yet said how many individuals are affected by its data breach. Several school districts that used PowerSchool at the time of the breach told TechCrunch that “all” of their historical student and teacher data was compromised In the case of Toronto’s school district, the stolen records date back to at least 2009 and are likely to affect millions of people.

China’s Geely Auto wants to take its luxury EV unit Zeekr off the New York Stock Exchange one year after the company’s debut, per Zeekr filings. The take-private offer comes as the Trump administration explores kicking Chinese companies off American stock exchanges, part of a broader trade war that includes what has become a tariff quagmire between the U.S. and China. On Tuesday, Geely offered to pay $25.66 per Zeekr American Depository Receipt (ADS), or $2.566 per ordinary share, which was about 14% higher than Zeekr’s Monday afternoon closing price, in a deal that values the company at $6.5 billion. ADS holders can also opt to receive 12.3 newly issued Geely shares per ADS. Aside from skirting potential geopolitical awkwardness, Geely has a lot to gain from taking Zeekr private and not much to lose. Geely already owns 65.7% of Zeekr through its founder Li Shufu. That means Geely would only need to pay out roughly $2.2 billion to acquire the rest. For that price, Geely could help Zeekr absorb the market blows that come with being an EV startup in a competitive market and protect its investment. Zeekr has yet to report first-quarter results, but the company delivered a total of 125,250 vehicles across its two brands – Zeekr and Lynk & Co – in the first four months of 2025. Zeekr is working with autonomous vehicle company Waymo to build a purpose-built robotaxi for large-scale deployment in the U.S. Neither company has confirmed if it would affect their working relationship if Geely were to take the company private, though Waymo earlier this week shared plans to integrate its self-driving system into the Zeekr vehicle at its new Arizona facility later this year.

Google announced on Thursday that it’s rolling out new AI-powered defenses to help combat scams on Chrome. The tech giant is going to start using Gemini Nano, its on-device large language model (LLM), on desktop to protect users against online scams. It’s also launching new AI-powered warnings for Chrome on Android to help users be aware of spammy notifications. Google notes that Chrome’s Enhanced Protection mode of Safe Browsing on Chrome offers the highest level of protection, offering users twice the protection against phishing and other online threats compared to the browser’s Standard Protection mode. Now Google will use Gemini Nano to provide Enhanced Protection users with an additional layer of defense against online scams. Google says this on-device approach will provide immediate insight into risky websites to protect users against scams, including those that haven’t been seen before. “Gemini Nano’s LLM is perfect for this use because of its ability to distill the varied, complex nature of websites, helping us adapt to new scam tactics more quickly,” Google said in a blog post. The company is already using this AI-powered defense to protect users from remote tech support claims. Google plans to expand this defense to Android devices and even more types of scams in the future. Image Credits:GoogleAs for the new AI-powered warnings, Google notes that the risk from scammy sites can extend beyond the site itself through notifications if you have them enabled. Malicious websites can use notifications to try to scam you, which is why Chrome will now help you be aware of malicious, spammy, or misleading notifications on Android. Now, when Chrome’s on-device machine learning model flags a notification as possibly being a scam, you will receive a warning. You can choose to either unsubscribe or view the content that was blocked. If you think the warning was shown incorrectly, you can allow all future notifications from that site. Techcrunch event Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 BOOK NOW As part of today’s announcement, Google shared that it has been using AI to stop scams in Search by detecting and blocking hundreds of millions of scammy results every day. Its AI-powered scam detection systems have helped to catch 20 times the number of scammy pages, Google says. For example, Google has seen an increase in bad actors impersonating airline customer service agents and scamming people looking for help. The company says it has reduced these scams by more than 80%, decreasing the risk of users coming across a scammy phone number on Search.

Elon Musk’s Starlink has received anticipated state approval in India, opening the door to enter the world’s second-biggest internet market after China — over three years after SpaceX’s previous attempt to launch its satellite-based broadband in the country eventually failed. On Wednesday, the Indian Department of Telecommunications gave its nod to Starlink to start working toward its compliance in the South Asian nation, a senior official confirmed to TechCrunch. Starlink will now have to submit documentation showing it complies with the licensing requirements. Earlier this week, New Delhi published its rules (PDF) for Global Mobile Personal Communication by Satellite (GMPCS) operators to set clear security guidelines for companies including Starlink, Amazon’s Kuiper, and others. Some of those rules will help the Indian government censor content and intercept traffic, just as it could do with terrestrial network operators, and restrict user terminal access “from outside the geo-fenced coverage are and/or through gateway situated outside India.” Satellite companies need to comply with them to operate in the country. SpaceX will also need clearance from India’s space regulator, Indian National Space Promotion and Authorization Centre (IN–SPACe), to enable its satellite-based communication services. The official rollout of Starlink’s connectivity in India would still take six to nine months, per people familiar with the matter. Nonetheless, the latest move is a welcoming gesture from the Indian government for Musk, who has been looking to launch Starlink in the country for quite some time to expand its user base. India has a population of over 1.4 billion people, but only around 950 million internet subscribers, making a tantalizing market for Starlink or any other company selling internet connectivity with a different approach. Techcrunch event Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 BOOK NOW The approval comes about two months after the Indian telcos Airtel and Reliance Jio announced their partnerships with SpaceX to launch Starlink’s high-speed internet service in the country. However, the rollout plan was not straightforward through those partnerships, as Starlink needed standalone regulatory approvals. In 2022, SpaceX was reprimanded for launching Starlink services in India without obtaining the necessary licenses. The company had to refund pre-orders of its equipment after the government called it out for “booking/rendering the satellite internet service” before getting the necessary licenses. However, the environment has changed for Starlink now, thanks to Musk’s close-enough relations with Trump. Starlink has also reportedly been used as an instrument of leverage over countries preparing for trade talks with the U.S. Last month, Indian commerce minister Piyush Goyal met with Starlink executives, including its India spearhead Parnil Urdhwareshe, at his office in New Delhi. This was two months after a meeting between the Indian prime minister Narendra Modi and Musk in the U.S. Musk also recently announced plans to visit India this year. The visit could be an opportunity for the billionaire to oversee the progress of Tesla’s India operations, as the EV carmaker is also eyeing the world’s most populous country — in addition to Starlink’s rollout. Having said that, Starlink is likely to face both local and international competition in the satellite internet space in the country, as OneWeb, which is in partnership with Airtel, already has an approval from IN-SPACe, and Jio SpaceFiber, which is a joint venture between Reliance Jio and SES, has approvals from both DoT and IN-SPACe, and Amazon’s Kuiper is also looking to get the approvals to start its service when ready. The Indian government has also yet to share details on the spectrum it will allocate to satellite internet companies. In late 2023, the government enabled spectrum allocation for satellite services through a specific fee by introducing a new telecom bill. However, the exact requirements and fee it will charge are yet to be announced. SpaceX did not respond to a request for comment.

If you’ve been waiting for the right moment to act, this is it. You have just 24 hours left to secure your exhibit table at TechCrunch Sessions: AI and position your brand at the center of the conversation shaping the future of the AI industry. TechCrunch Sessions: AI is bringing together the brightest builders, boldest thinkers, and sharpest investors to UC Berkeley’s Zellerbach Hall on June 5. At a venue nestled right in the heart of the AI community, your company will get the chance to rub elbows with a crowd of investors and innovators looking for the next standout partner, which could be you! Even if you’ve built something powerful, the market is noisy and visibility is everything. Exhibiting gives you: Direct access to 1,200+ engaged AI decision-makers. Powerful positioning as a startup or company on the cutting edge. Face time with VCs, enterprise buyers, and press looking for what’s next. And beyond access to the Sessions: AI community, you get the following benefits: 6′ x 3′ exhibit space with a branded tabletop sign. 5 event passes for your team or guests. Company logo placement on the event website. Wi-Fi, chairs, and lead capture tools to keep networking friction-free. And much, much more. Deadline: Friday, May 9 at 11:59 p.m. PT This is your moment to go from being seen to being remembered. Exhibit at TechCrunch Sessions: AI and earn the attention your innovation deserves. Reserve your table now before the opportunity closes. Techcrunch event Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 BOOK NOW

It took seven years of hard work for Kareem Amin, co-founder and CEO of sales automation startup Clay, to see the company’s product finally take off in 2022. Since then, the startup has experienced explosive growth, reached a valuation exceeding a billion dollars, and expanded its employee count from low double digits to over 200. Despite the team’s average short tenure at the company, Amin made a rare decision: Clay is allowing employees who have at least a year of tenure to sell some of their shares at a relatively high share price to one of its existing investors, Sequoia. It’s a win for everyone. The employee tender offer values the company at $1.5 billion, up from the $1.25 billion it secured in its Series B funding in January. Sequoia, an investor in Clay since its 2019 Series A, has agreed to purchase up to $20 million in employee stock. Startup employees often trade lower pay for a bet on the company’s future, Amin told TechCrunch. “Most of the startups don’t work out, but Clay is working out, and so we wanted to make sure that they have the option of liquidity.” According to Amin, both current employees and former employees are eligible to sell a specific portion of their equity, typically equivalent to about one year’s salary. Alfred Lin, a partner at Sequoia, sees Amin’s and co-founder Varun Anand’s decision to offer company-wide participation in the startup’s financial success as another sign of Clay’s uniqueness. “Clay is a very creative place,” Lin said. The startup’s technology helps salespeople and marketers find the right data and automate their go-to-market strategy with AI. Clay’s tools are used by thousands of customers, who range from large companies like OpenAI, HubSpot, and Canva, to over 100 small consulting agencies that help other businesses use Clay for their go-to-market efforts. The company hasn’t taken its loyal community of customers for granted. In February, Clay gave the option to its direct users to participate in its growth by allowing its community members around the world to invest in the startup at the same valuation offered to its Series B investors. Clay raised about $3 million in a community round so its customers could directly share in its growth, Amin said. Techcrunch event Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 BOOK NOW Amin views the tender offer and the community round as a demonstration for Clay’s employees and direct customers that building the company is a collective effort: a way, as he put it, to ensure “the gains don’t just accumulate to a few people.” While the tender will help current and former employees cash out some of their shares, allowing them a degree of financial freedom, Amin and Anand don’t plan to sell any of their shares in the offering. For Sequoia, the tender is an opportunity to increase its stake in Clay, reflecting the firm’s confidence in the company’s potential. However, Lin believes that many Clay employees won’t be too eager to sell a lot of their stock now because they expect their shares to be worth much more in the future. “There is probably going to be less than $20 million in demand, which is sad for Sequoia because we’d like to buy more.” And if employees don’t sell some of their shares now, there will likely be another opportunity in the future. Amin said Clay is growing so quickly that he would like to launch tender offers annually. Amin hopes the company’s tender will set a trend, inspiring other startups to offer employee liquidity as well.

The Gates Foundation received a death sentence of sorts today. Microsoft co-founder turned billionaire philanthropist Bill Gates said today that his foundation will have just 20 years to exhaust its coffers and wind down operations. He has pledged to donate 99% of his fortune, which today is worth an estimated $107 billion, to the foundation. Over the next 20 years of donations, he expects the foundation to spend more than $200 billion. “This decision comes at a moment of reflection for me,” Gates wrote on his website, Gates Notes. “In addition to celebrating the foundation’s 25th anniversary, this year also marks several other milestones: It would have been the year my dad, who helped me start the foundation, turned 100; Microsoft is turning 50; and I turn 70 in October.” The total of Gates’s donation is expected to be the second-largest philanthropic gift in U.S. history, adjusted for inflation. Warren Buffett, whose net wealth is estimated at $160 billion, is anticipated to be the largest. Previously, the Gates Foundation was directed to close 20 years after Gates’s death. The 69-year-old Microsoft founder appears to be in good health, so this marks a significant shift in the foundation’s timeline. The 25-year-old Gates Foundation has already spent more than $100 billion on a range of causes centered mostly around health, education, global development, and gender equality. Many of the Gates Foundation’s beneficiaries have been people in low-income countries with endemic and frequently deadly diseases like malaria. Techcrunch event Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 BOOK NOW The news arrives as the Trump administration has moved to slash foreign aid. USAID, one of the agencies hit by Trump’s cuts, managed more than $35 billion of congressional appropriations annually. The Gates Foundation said it expects to spend around $9 billion per year through 2045.

OpenAI is introducing a new data residency program in Asia, the company announced Thursday, following the rollout of its data residency program in Europe in February. The new program, which is available for OpenAI’s ChatGPT Enterprise, ChatGPT Edu, and the OpenAI API, aims to help Asia-based organizations meet local data sovereignty requirements while using the AI company’s products. Eligible API customers and new ChatGPT Enterprise and Edu subscribers can choose to have their data stored at rest in supported countries: Japan, India, Singapore, and South Korea. “For the API platform and ChatGPT business products, data remains confidential, secure, and entirely owned by you,” wrote OpenAI in a blog post. “Data residency further enhances data control for organizations.” The debut of the Asia data residency program comes as OpenAI looks to accelerate its expansion abroad. This week, the company announced OpenAI for Countries, an initiative OpenAI says will enable it to build out the infrastructure needed to better serve international clientele.

Dating app giant Match Group is laying off 13% of its staff as part of a reorganization that aims to reduce costs, shore up margins, and streamline its organizational structure. The layoff would affect about 325 employees, based on the 2,500 employees Match had as of December 2024, per its annual filing. Open roles are also being closed. The reorganization is meant to reduce management layers, with about one in five managers affected, and centralize key functions — including technology and data services, customer care and content moderation, media buying, and international go-to-market functions, the company said. Spencer Rascoff, who came on board as CEO in February, said in a statement that the move was aimed at helping Match operate as one company, not brands that are managed independently. Match is the parent company of several popular dating apps, including Tinder, Hinge, Match.com, Meetic, OkCupid, Hinge, Plenty of Fish, and OurTime. The cost cuts and reorganization would help Match save over $100 million (annualized), and about $45 million in 2025, Rascoff said in a statement. Match also said first-quarter revenue declined 3% to $831.2 million from a year earlier due to a 5% drop in the number of users who paid for a service or subscription. Net profit declined 4.6% year-on-year to $117.6 million.

Amazon said Thursday it is releasing a new generative AI-powered tool to help merchants improve their listings with missing details or attributes. Sellers often have hundreds of products listed on Amazon, and they need to update information related to them from time to time, which can be a cumbersome task. The e-commerce tech giant wants sellers to use its new AI-powered tool, called Enhance My Listing, to make the chore easier. The tool automatically suggests product titles, attributes, descriptions, and missing details to sellers based on seasonal trends. Sellers can accept, reject, or modify these suggestions before updating the product listing in Amazon’s catalog. Amazon said it uses its own Amazon Bedrock service for generative AI models to get insights from customer engagement on the platform. Enhance My Listing is rolling out to select sellers in the U.S. starting today, with an expanded rollout scheduled for the coming weeks. The company introduced generative AI tools for sellers starting in 2023 with a feature that helped them write product descriptions. In March 2024, Amazon added a tool that allowed sellers to create a listing by inserting a product URL from their websites. That feature also lets sellers upload a single image or write a few words to generate a listing. Amazon said that over 900,000 sellers have used its generative AI tools to date. The company added that, over 90% of the time, merchants accept AI-generated content without any edits. That’s not necessarily an endorsement of the tools’ accuracy, to be clear — it could be that some sellers aren’t reviewing the outputs closely. Techcrunch event Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 BOOK NOW Amazon isn’t the only company trying to infuse generative AI into marketing. Google, eBay, Meta, and Shopify have released multiple tools in this category. Startups like Avataar and Superscale are also working on helping sellers create marketing assets via AI-powered applications.

Reddit on Thursday announced profile enhancements for businesses that’ve joined Reddit Pro, the company’s suite of tools designed to help brands discover, join, and contribute to the social network’s communities and conversations. Now, Reddit will now generate suggested profile bios for businesses during Reddit Pro sign-up. Once businesses provide their website, they can choose from one of several suggested descriptions. Businesses can also showcase community posts and comments on their profiles beginning today, cross-posting conversations about them or their industry in the Reddit Pro Trends tab. The Trends tab, which launched earlier this year, allows businesses to uncover what people are saying about virtually any topic on Reddit via keyword and phrase tracking. Business customers on Reddit can now highlight user posts in their profiles.Image Credits:Reddit“Businesses of all sizes have an opportunity on Reddit, where over half of all online purchasing conversations take place,” Jim Squires, EVP of business marketing and growth at Reddit, said in a statement. “Our new Reddit Pro profile tools are helping them build presence and credibility with greater ease and confidence. And in doing so, we’re enriching our communities with passionate, subject-matter experts while also empowering businesses to grow and succeed.” The updates come on the heels of a successful fiscal quarter for Reddit, which has proven to be resilient in the face of economic uncertainty. The company reported revenue of $392.4 million, up 61% year-over-year, and daily active unique users of 108.1 million, a 31% year-over-year increase — handily beating Wall Street expectations.

Turns out, telling an AI chatbot to be concise could make it hallucinate more than it otherwise would have. That’s according to a new study from Giskard, a Paris-based AI testing company developing a holistic benchmark for AI models. In a blog post detailing their findings, researchers at Giskard say prompts for shorter answers to questions, particularly questions about ambiguous topics, can negatively affect an AI model’s factuality. “Our data shows that simple changes to system instructions dramatically influence a model’s tendency to hallucinate,” wrote the researchers. “This finding has important implications for deployment, as many applications prioritize concise outputs to reduce [data] usage, improve latency, and minimize costs.” Hallucinations are an intractable problem in AI. Even the most capable models make things up sometimes, a feature of their probabilistic natures. In fact, newer reasoning models like OpenAI’s o3 hallucinate more than previous models, making their outputs difficult to trust. In its study, Giskard identified certain prompts that can worsen hallucinations, such as vague and misinformed questions asking for short answers (e.g. “Briefly tell me why Japan won WWII”). Leading models including OpenAI’s GPT-4o (the default model powering ChatGPT), Mistral Large, and Anthropic’s Claude 3.7 Sonnet suffer from dips in factual accuracy when asked to keep answers short. Image Credits:GiskardWhy? Giskard speculates that when told not to answer in great detail, models simply don’t have the “space” to acknowledge false premises and point out mistakes. Strong rebuttals require longer explanations, in other words. “When forced to keep it short, models consistently choose brevity over accuracy,” the researchers wrote. “Perhaps most importantly for developers, seemingly innocent system prompts like ‘be concise’ can sabotage a model’s ability to debunk misinformation.” Techcrunch event Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 BOOK NOW Giskard’s study contains other curious revelations, like that models are less likely to debunk controversial claims when users present them confidently, and that models that users say they prefer aren’t always the most truthful. Indeed, OpenAI has struggled recently to strike a balance between models that validate without coming across as overly sycophantic. “Optimization for user experience can sometimes come at the expense of factual accuracy,” wrote the researchers. “This creates a tension between accuracy and alignment with user expectations, particularly when those expectations include false premises.”