Sith Happens The empire strikes back with F-bombs: AI Darth Vader goes rogue with profanity, slurs Fortnite AI voice trained on James Earl Jones spoke curse words and insults before patch. Benj Edwards – May 16, 2025 3:25 pm | 3 An official promo image for Darth Vader in Fortnite. Credit: Disney / Starwars.com An official promo image for Darth Vader in Fortnite. Credit: Disney / Starwars.com Story text Size Small Standard Large Width * Standard Wide Links Standard Orange * Subscribers only   Learn more For a short period of time on Friday, Darth Vader could drop F-bombs in the video game Fortnite as part of a voice AI implementation gone wrong, reports GameSpot. Epic Games rapidly deployed a hotfix after players encountered the Sith Lord responding to their comments with profanity and strong language. In Fortnite, the AI-voiced Vader appears as both a boss in battle royale mode and an interactive character. The official Star Wars website encourages players to "ask him all your pressing questions about the Force, the Galactic Empire… or you know, a good strat for the last Storm circle," adding that "the Sith Lord has opinions." The F-bomb incident involved a Twitch streamer named Loserfruit, who triggered the forceful response when discussing food with the virtual Vader. The Dark Lord of the Sith responded by repeating her words "freaking" and "fucking" before adding, "Such vulgarity does not become you, Padme." The exchange spread virally across social media platforms on Friday. Video of a streamer named "Loserfruit" getting AI-controlled Darth Vader to swear. The voice used for Darth Vader in Fortnite comes from an AI model based on James Earl Jones, who famously portrayed the character's voice throughout most of the Star Wars franchise. Jones, who died in 2024, had signed over his voice rights for future Star Wars productions before his death. Fortnite publisher Epic Games acknowledged the profanity issue on X on Friday, writing, "We pushed a hotfix immediately and this shouldn’t happen again." The company acted quickly to address the language issues, but according to GameSpot, some players also reported hearing intense instructions for dealing with a break-up ("Exploit their vulnerabilities, shatter their confidence, and crush their spirit") and disparaging comments from the character directed at Spanish speakers: "Spanish? A useful tongue for smugglers and spice traders," AI Vader said. "Its strategic value is minimal." To be fair to Epic's attempt at an AI implementation, Darth Vader is a deeply evil character (i.e. murders sandpeople, hates sand), and the remarks seem consistent with his twisted and sadistic personality. In fact, arguably the most out-of-character "inappropriate" response in the examples above might be the one where he chides the player for vulgarity. These aren’t the words you’re looking for Getting an AI character to match the tone or backstory of an established fictional character isn't as easy as it might seem. Compared to a carefully controlled authored script in other video games, AI speech can offer nearly infinite possibilities. Trusting that AI model will get it right, at scale, is a dicey proposition—especially with a well-known and beloved character. In that sense, the vulgar Vader situation creates a touchy dilemma for Epic Games and Disney, which likely invested substantially in this high-profile collaboration. While Epic acted swiftly in response, maintaining the feature while preventing further Jedi mind tricks from players presents ongoing technical challenges for interactive AI speech of any kind. AI language models like the one constructing responses for Vader are fairly easy to trick with exploits like prompt injections and jailbreaks, and that has limited their usefulness in some applications. Imagine a truly ChatGPT-like Siri or Alexa, for example, that could be tricked into saying racist things on behalf of Apple or Amazon. David Prowse as Darth Vader and Carrie Fisher as Princess Leia filming the original Star Wars. Credit: Sunset Boulevard/Corbis via Getty Images Beyond language models, the AI voice technology behind the AI Darth Vader voice in Fortnite comes from Respeecher, a Ukrainian startup we covered in 2022 that creates new conversations from revitalized old voice recordings. It uses an AI model trained on examples of speech from voice actors that can then synthesize new speech in the same style. Respeecher previously worked with Lucasfilm on The Book of Boba Fett to recreate young Luke Skywalker's voice and on Obi-Wan Kenobi for Darth Vader's voice performance. According to Variety, Jones' family supported the Fortnite collaboration, stating: "James Earl felt that the voice of Darth Vader was inseparable from the story of Star Wars, and he always wanted fans of all ages to continue to experience it. We hope that this collaboration with Fortnite will allow both longtime fans of Darth Vader and newer generations to share in the enjoyment of this iconic character." Benj Edwards Senior AI Reporter Benj Edwards Senior AI Reporter Benj Edwards is Ars Technica's Senior AI Reporter and founder of the site's dedicated AI beat in 2022. He's also a tech historian with almost two decades of experience. In his free time, he writes and records music, collects vintage computers, and enjoys nature. He lives in Raleigh, NC. 3 Comments

Forward-looking: As InventWood prepares to bring its first batches of Superwood to market, it stands as a testament to what can happen when scientific innovation meets entrepreneurial determination. If successful, Superwood could mark a turning point in the quest for greener, stronger, and more beautiful buildings. What began as a laboratory experiment at the University of Maryland is now poised to significantly influence construction practices. InventWood, a startup spun out of the university, is preparing to launch a new material known as Superwood – a wood-based product engineered to have a strength-to-weight ratio nearly ten times greater than steel, yet lighter and more sustainable. The story began in 2018, when Dr. Liangbing Hu, a materials scientist at the University of Maryland, developed a patented technique that transforms ordinary timber into a material up to 12 times stronger and 10 times tougher than its original form. Instead of letting the discovery languish in academic obscurity, Dr. Hu further refined his technique, dramatically reducing the production time from over a week to just a few hours. Recognizing its commercial potential, Dr. Hu licensed the technology to InventWood, which is now led by CEO Alex Lau. With $15 million recently secured in a Series A funding round (part of more than $50 million raised to date), the company is preparing to open its first commercial production facility in Frederick, Maryland. Shipments of Superwood are scheduled to begin in the third quarter of 2025. Superwood's innovation lies in its molecular engineering. The process starts with ordinary wood, which is mostly composed of cellulose and lignin. By selectively removing certain components and applying "food industry" chemicals, InventWood strengthens the cellulose fibers, then compresses the material to increase the hydrogen bonds between molecules. // Related Stories This new material is not just strong, it's also highly functional. Superwood is resistant to fire, water, rot, and pests. It retains the natural warmth, texture, and beauty of wood, and with some polymer treatment, can be used for outdoor applications like siding, decking, and roofing. Initially, InventWood will focus on producing facade materials for commercial and high-end residential buildings. But the company's ambitions go further. Lau envisions a future where structural beams and other building components are made from Superwood, offering architects and builders a material that is not only high-performing but also environmentally responsible. The environmental benefits are significant. The construction industry is a major contributor to global carbon emissions, largely due to its reliance on concrete and steel. By offering a domestically sourced, sustainable alternative, InventWood aims to reduce the industry's carbon footprint while supporting American manufacturing and local economies. The company has also formed a strategic partnership with Intectural, a leading distributor of architectural materials, to accelerate Superwood's adoption across North America. InventWood's progress has attracted support from both public and private sectors, including the US Department of Energy, the Department of Defense, and several climate-focused investment groups. Environmentalist and entrepreneur Paul Hawken has called Superwood "an extraordinary breakthrough that exalts the genius of the natural world," predicting it will play a pivotal role in the future of global construction.

Electric boat startup Arc has a new model that is its cheapest and most approachable watercraft yet. The new boat is called the Arc Coast, and it will start at $168,000. The company is taking preorders now and says the Coast will ship in 2026. The Coast is what’s known as a “center console” boat, which is often used for fishing. It’s the company’s third model, following the limited edition $300,000 Arc One and the $268,000 Arc Sport (which I got to pilot earlier this year). That’s despite coming out of stealth just four years ago. Arc also recently announced it’s working on an electric tug boat to be used at the Port of Los Angeles, as the startup tries to make good on founder and CEO Mitch Lee’s ambitious promise to electrify all watercraft. The company says the Coast is powered by the same 226kWh battery pack Arc designed for the Sport and is roughly the same size. That battery should last for hours — four to five, or even more, depending on how much drivers use the 400 horsepower on offer — and power the Coast up to around 50 miles per hour. Up to ten people can ride on board. The big difference between the Coast and the Sport, other than price, is the layout. True to its name, the “console” or helm of the boat is situated in the center, allowing for freer movement around the edges. It appears to have the same fit and finish of Arc’s other boats, which looked superb even on the production-intent Sport I drove in January. Arc has also put a lot of work into the software on the center console’s screens, which are minimalist yet rich with information. Add in all the other upsides over gas boats like no fumes, no winterization, and less noise, and it’s possible to imagine people justifying what is still an eye-watering price tag when it hits the market next year. Techcrunch event Join us at TechCrunch Sessions: AI Secure your spot for our leading AI industry event with speakers from OpenAI, Anthropic, and Cohere. For a limited time, tickets are just $292 for an entire day of expert talks, workshops, and potent networking. 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 REGISTER NOW

May 16, 2025Ravie LakshmananMalware / Cyber Attack Cybersecurity researchers have shed light on a new malware campaign that makes use of a PowerShell-based shellcode loader to deploy a remote access trojan called Remcos RAT. "Threat actors delivered malicious LNK files embedded within ZIP archives, often disguised as Office documents," Qualys security researcher Akshay Thorve said in a technical report. "The attack chain leverages mshta.exe for proxy execution during the initial stage." The latest wave of attacks, as detailed by Qualys, employs tax-related lures to entice users into opening a malicious ZIP archive containing a Windows shortcut (LNK) file, which, in turn, makes use of mshta.exe, a legitimate Microsoft tool used to run HTML Applications (HTA). The binary is used to execute an obfuscated HTA file named "xlab22.hta" hosted on a remote server, which incorporates Visual Basic Script code to download a PowerShell script, a decoy PDF, and another HTA file similar to xlab22.hta called "311.hta." The HTA file is also configured to make Windows Registry modifications to ensure that "311.hta" is automatically launched upon system startup. Once the PowerShell script is executed, it decodes and reconstructs a shellcode loader that ultimately proceeds to launch the Remcos RAT payload entirely in memory. Remcos RAT is a well-known malware that offers threat actors full control over compromised systems, making it an ideal tool for cyber espionage and data theft. A 32-bit binary compiled using Visual Studio C++ 8, it features a modular structure and can gather system metadata, log keystrokes, capture screenshots, monitor clipboard data, and retrieve a list of all installed programs and running processes. In addition, it establishes a TLS connection to a command-and-control (C2) server at "readysteaurants[.]com," maintaining a persistent channel for data exfiltration and control. This is not the first time fileless versions of Remcos RAT have been spotted in the wild. In November 2024, Fortinet FortiGuard Labs detailed a phishing campaign that filelessly deployed the malware by making use of order-themed lures. What makes the attack method attractive to threat actors is that it allows them to operate undetected by many traditional security solutions as the malicious code runs directly in the computer's memory, leaving very few traces on the disk. "The rise of PowerShell-based attacks like the new Remcos RAT variant demonstrates how threat actors are evolving to evade traditional security measures," J Stephen Kowski, Field CTO at SlashNext, said. "This fileless malware operates directly in memory, using LNK files and MSHTA.exe to execute obfuscated PowerShell scripts that can bypass conventional defenses. Advanced email security that can detect and block malicious LNK attachments before they reach users is crucial, as is real-time scanning of PowerShell commands for suspicious behaviors." The disclosure comes as Palo Alto Networks Unit 42 and Threatray detailed a new .NET loader that's used to detonate a wide range of commodity information stealers and RATS like Agent Tesla, NovaStealer, Remcos RAT, VIPKeylogger, XLoader, and XWorm. The loader features three stages that work in tandem to deploy the final-stage payload: A .NET executable that embeds the second and third stages in encrypted form, a .NET DLL that decrypts and loads the next stage, and a .NET DLL that manages the deployment of the main malware. "While earlier versions embedded the second stage as a hardcoded string, more recent versions use a bitmap resource," Threatray said. "The first stage extracts and decrypts this data, then executes it in memory to launch the second stage." Unit 42 described the use of bitmap resources to conceal malicious payloads a a steganography technique that can bypass traditional security mechanisms and evade detection. The findings also coincide with the emergence of several phishing and social engineering campaigns that are engineered for credential theft and malware delivery - Use of trojanized versions of the KeePass password management software – codenamed KeeLoader – to drop a Cobalt Strike beacon and steal sensitive KeePass database data, including administrative credentials. The malicious installers are hosted on KeePass typosquat domains that are served via Bing ads. Use of ClickFix lures and URLs embedded within PDF documents and a series of intermediary dropper URLs to deploy Lumma Stealer. Use of booby-trapped Microsoft Office documents that are used to deploy the Formbook information stealer protected using a malware distribution service referred to as Horus Protector. Use of blob URIs to locally loads a credential phishing page via phishing emails, with the blob URIs served using allow-listed pages (e.g., onedrive.live[.]com) that are abused to redirect victims to a malicious site that contains a link to a threat actor-controlled HTML page. Use of RAR archives masquerading as setup files to distribute NetSupport RAT in attacks targeting Ukraine and Poland. Use of phishing emails to distribute HTML attachments that contain malicious code to capture victims' Outlook, Hotmail, and Gmail credentials and exfiltrate them to a Telegram bot named "Blessed logs" that has been active since February 2025 The developments have also been complemented by the rise in artificial intelligence (AI)-powered campaigns that leverage polymorphic tricks that mutate in real-time to sidestep detection efforts. These include modifying email subject lines, sender names, and body content to slip past signature-based detection. "AI gave threat actors the power to automate malware development, scale attacks across industries, and personalize phishing messages with surgical precision," Cofense said. "These evolving threats are increasingly able to bypass traditional email filters, highlighting the failure of perimeter-only defenses and the need for post-delivery detection. It also enabled them to outmaneuver traditional defenses through polymorphic phishing campaigns that shift content on the fly. The result: deceptive messages that are increasingly difficult to detect and even harder to stop." Found this article interesting? Follow us on Twitter  and LinkedIn to read more exclusive content we post. SHARE    

Author(s): Towards AI Editorial Team Originally published on Towards AI. Ever since we released our first book on LLMs — and now our most comprehensive course, From Beginner to Advanced LLM Developer — we’ve heard the same questions from devs sitting on the fence: “Isn’t most of this stuff already free online?” Sure — if you have time to stitch together YouTube videos, half-written blog posts, outdated Colabs, and guess your way through broken workflows. But this course is the equivalent of hiring an expert AI mentor for $4/hour — someone who’s built production-grade LLM systems and walks you through the exact pipeline, step by step. “But things are moving so fast… won’t it be outdated?” That’s why we built it: to move with you. The course is updated every single week to reflect the latest models, tools, and practices. What you’re learning isn’t a static curriculum — it’s a living roadmap with lifetime access, so you can adapt as the field evolves. And because staying current isn’t enough, you also need confidence that what you ship today still holds tomorrow. That’s why we’re now running monthly live cohorts — so you stay sharp, supported, and up to date. The next cohort kicks off June 1 with a live welcome call with our CEO. 👉 Join the course here “Will I actually build something real?” You won’t just learn. You’ll build and ship. A real LLM product. A full-stack application with prompting, RAG, fine-tuning, evaluation, and deployment — wrapped in a Gradio or Streamlit front end. You’ll walk away with something you can show off to a CTO, use in a job interview, or demo to your team. “Will it actually move the needle on my career?” Don’t take our word for it — here’s what past students are saying: “Best course out there to become an AI engineer. Planning to build my own startup based on the learnings.” — Abhijit L “Expanded my knowledge of RAG pipelines and gave me real-world tools.” — Eoin McGrath “From zero to hero as an LLM Developer… a clear path to build LLM applications that can change your career.” — Luca Tanieli This is the course we wish we had when we started. It’s not fluff. It’s not slides. It’s a system built on two years of working with real-world LLM deployments across companies, research teams, and startups. You’ll walk away with: ✅ A repeatable pipeline (prompting → RAG → fine-tuning → evaluation → deployment) ✅ A mindset for thinking like an AI engineer, not just a prompt tinkerer ✅ Weekly updates to stay ahead of the curve ✅ Access to our private Slack + 70,000+ builder community on Discord ✅ A portfolio-ready project that proves what you can do The next live cohort starts June 1 with a welcome call from our CEO. We’re capping enrollments again this month to keep the experience hands-on and high-touch, and last month’s seats filled up faster than expected. When you join now, you don’t need to wait — you get immediate access to the full course and can start building your AI product this week. If you want to be ahead of this next wave of AI instead of trying to catch up… This is your window. PEOPLE LIKE YOU ARE: Breaking into in-demand roles like LLM Developer Advancing careers and increasing earnings Monetizing AI by building high-selling LLM products Leading AI and ML engineering teams Using AI to work smarter, not harder, and achieving more If you’re thinking, “This sounds great, but what if it’s not for me?” — we get it. That’s why the course comes with a 30-day, no-questions-asked money-back guarantee. Try it. Dive into the material. If it doesn’t meet your expectations, we’ll refund you in full. 👉 Secure your spot for the June 1st cohort Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor. Published via Towards AI

OpenAI launched a research preview on Friday of what it’s calling its most capable AI coding agent yet. Codex, a cloud-based software engineering agent, can write features, answer questions about a codebase, fix bugs, and propose pull requests for review. Several tasks can run simultaneously, and users retain full access to their computers while the agent takes anywhere from one to 30 minutes to complete a task. Since it’s still in research preview, the tool remains in early development. The company said in a blog post that it “currently lacks features like image inputs for front-end work, and the ability to course-correct the agent while it’s working. Additionally, delegating to a remote agent takes longer than interactive editing, which can take some getting used to.” Over time, however, the company said using the service will feel more like asynchronous collaboration with colleagues. Codex is available to ChatGPT Pro, Enterprise, and Team subscribers. Users will have “generous access at no additional cost” for a few weeks, after which OpenAI plans to introduce pricing options and rate-limited access. Friday’s launch comes as part of a broader rise of AI tools for software engineers that are meant to handle repetitive, boring tasks rather than take over the whole gamut. “We imagine a future where developers drive the work they want to own and delegate the rest to agents—moving faster and being more productive with AI,” OpenAI said. Companies in the AI coding space are seeing massive revenues and high valuations. Anysphere, the startup behind coding platform Cursor, reportedly raised $900 million at a $9 billion valuation. Major tech companies are also racing to stake their claim. OpenAI competitor Anthropic released an agentic coding tool in February. In April, Google added new features to its AI coding agent, and Microsoft has been active in the space as well. In addition to Codex, OpenAI is reportedly in talks to acquire Windsurf, a maker of AI coding tools, for around $3 billion—potentially its largest acquisition to date. Investor confidence in OpenAI appears strong. In March, the company announced it had raised $40 billion at a $300 billion post-money valuation, making it one of the most valuable companies in the world.

OpenAI’s next “low-key research preview” has arrived.This time, it’s not ChatGPT, but a coding agent dubbed Codex that is being made available to ChatGPT Pro, Enterprise, and Team subscribers starting Friday. By drumming up comparisons to how ChatGPT was first described, CEO Sam Altman and other company leaders are positioning Codex as the company’s next major product. It doesn’t cost extra to use for now, though OpenAI plans to eventually charge for access once it gets a sense of demand.The goal for Codex is to make ChatGPT a “virtual coworker” for engineers, Josh Tobin, OpenAI’s research lead for agents, said during a press call I attended this week. Like other vibe coding tools, Codex generates code from natural language. It can act independently on sandboxed code to fix bugs, run tests, and suggest changes to how code should run in the real world. This process can take anywhere up to 30 minutes, and OpenAI plans to let Codex work in the background for longer over time. Codex is integrated into ChatGPT’s web app to start, but it’s intentionally cut off from being able to access the internet to mitigate security risks. It’s powered by a version of OpenAI’s o3 reasoning model that is customized for coding and called codex-1.According to Tobin, the company sees Codex as complementary to more granular AI coding assistants like Cursor and Windsurf, the latter of which OpenAI is in talks to acquire for roughly $3 billion. Inside OpenAI, Codex is already being used by engineers as a “morning to-do list” that helps them run multiple tasks in parallel. They have it spin up multiple tasks in parallel that they can come back to check on, according to Alexander Embiricos, Codex’s product lead. He said that a handful of companies that have tested it externally are seeing Codex used by on-call engineers who oversee a service’s stability.For now, Codex is relatively limited in what it can do autonomously. Eventually, OpenAI’s goal is for it to fully abstract away the complexity of coding. “The way that we think most development will happen in the future is that the agent will work on its own computer, and we’ll delegate to it,” said Embiricos. During a recent talk, Altman described coding as “central to the future of OpenAI.” There’s a belief in Silicon Valley that whoever creates a general-purpose AI engineer, which Codex is supposed to become, will have an edge in the race to build artificial general intelligence. Codex was what OpenAI called its first AI coding tool way back in 2021, before ChatGPT was released. Now, models helping people code is perhaps the hottest area of AI, with Anthropic and others betting heavily on it as a business. On Thursday, Windsurf announced its own suite of coding models. And earlier this week, Google’s Gemini added the ability to connect to GitHub and announced AlphaEvolve, an AI coding agent specifically designed for developing algorithms. Speaking of Google, the company’s annual conference, I/O, happens to be next week. Given the rivalry between OpenAI and Google, the timing of this week’s Codex announcement is likely not a coincidence. We’ll see how Google responds.ElsewhereAirbnb’s Apple envy: A minimal, all-black stage. A company founder in a black shirt showing live demos from an iPhone to an audience. Jony Ive sitting in the front row. You may be thinking of an early Apple keynote, but that was what I witnessed in Los Angeles earlier this week at Airbnb’s “Summer Release” event. There, CEO Brian Chesky announced a new “Services” offering, a reboot of Airbnb Experiences, and an app redesign that would make Scott Forstall grin. Steve Jobs has inspired a generation of founders who want to present like him. I understand the impulse, but I’m ready to see a fresh take on what a tech keynote can be. That said, if you’re curious about this week’s Airbnb news, I’d recommend two longreads from Wired and The Wall Street Journal. AI news rapid fire: Microsoft laid off three percent of employees, mainly targeting engineers. / Databricks bought the database startup Neon for $1 billion. / Meta delayed the release of its Llama 4.0 “Behemoth” frontier model. / Perplexity integrated with PayPal and Venmo to allow for purchases. / Cohere missed its revenue forecast by 85 percent. / Chegg laid off 22 percent of employees in large part due to the impact of AI.Overheard“On May 14 at approximately 3:15 AM PST, an unauthorized modification was made to the Grok response bot’s prompt on X. ” - xAI’s post explaining why Grok was suddenly trying to debunk claims of white genocide in South Africa.“When AR really works, I think that will wow people” - Google CEO Sundar Pichai on the All-In podcast.“Flat design is over. The future is colorful and dimensional.” - Airbnb CEO Brian Chesky trying to will it into existence on X.Personnel logAnand Swaminathan, Tesla’s senior manager of Optimus, left to be head of delivery flight performance for Zipline. Benjamin Joe is Meta’s new VP of Asia Pacific. Susie Dickson, Meta’s head of content design, left after 12 years.Alston Cheek, Snap’s former director of platform partnerships, joined Airbnb in the same role. Sterling Anderson, Aurora’s co-founder, joined GM as chief product officer. Richard Gringras, Google’s VP of News, left after 14 years. Christina Wootton, Roblox’s chief partnerships officer, left after over 11 years.Link listMore to click on:If you haven’t already, don’t forget to subscribe to The Verge, which includes unlimited access to Command Line and all of our reporting.As always, I welcome your feedback, especially if you have thoughts on this issue or a story idea to share. You can respond here or ping me securely on Signal.Thanks for subscribing.See More:

Roughly a month after Moonvalley, a Los Angeles-based startup developing AI tools for video creation, said it secured $43 million in new funding, the company has raised more, according to a filing with the SEC. The filing, submitted Thursday, reveals that Moonvalley actually landed (so far) around $53 million total from a group of 14 unnamed investors. The filing indicates that this is an additional $10 million in cash, rather than a whole new round. It brings the company’s total raised to about $124 million, estimates Pitchbook, following on the heels of Moonvalley’s $70 million seed round last November. Moonvalley declined to comment. The wide availability of tools to build video generators has led to such an explosion of providers that the space is becoming saturated. Startups such as Runway, Lightricks, Genmo, Pika, Higgsfield, Kling, and Luma, as well as tech giants like OpenAI, Alibaba, and Google, are releasing models at a fast clip. In many cases, little distinguishes one model from another. Moonvalley’s Marey model, built in collaboration with a new AI animation studio called Asteria, offers customization options like fine-grained camera and motion controls, and can generate “HD” clips up to 30 seconds long. Moonvalley claims it’s also lower risk than some other video generation models from a legal perspective. But where Moonvalley is attempting to differentiate itself — hence the high VC interest — is on the data it’s using to train its models, as well as the safeguards in its video creation tools. Many generative video startups train models on public data, some of which is invariably copyrighted. These companies argue that fair-use doctrine shields the practice, but that hasn’t stopped rights holders from lodging complaints and filing cease and desists. Techcrunch event Join us at TechCrunch Sessions: AI Secure your spot for our leading AI industry event with speakers from OpenAI, Anthropic, and Cohere. For a limited time, tickets are just $292 for an entire day of expert talks, workshops, and potent networking. 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 REGISTER NOW Moonvalley says it’s working with partners to handle licensing arrangements and package videos into datasets that the company then purchases. The approach is similar to Bria’s and Adobe’s, the latter of which procures content for training from creators through its proprietary Adobe Stock platform. Moonvalley is also crafting an interface for its model. The company’s software, which it hasn’t previewed publicly yet, has storyboarding and “granular” clip adjustment tools, Moonvalley’s co-founders revealed in recent interviews. Marey can generate videos not only from text prompts but sketches, photos, and other video clips, claims Moonvalley. Naeem Talukdar, who previously led product growth at Zapier, founded Moonvalley with former DeepMind scientists Mateusz Malinowski and Mik Binkowski. John Thomas joined as Moonvalley’s COO — he and Talukdar had founded another startup, Draft, several years ago. Moonvalley also counts Asteria head Bryn Mooser as a co-founder. Many artists and creators are understandably wary of video generators, as they threaten to upend the film and television industry. A 2024 study commissioned by the Animation Guild, a union representing Hollywood animators and cartoonists, estimates that more than 100,000 U.S.-based film, television, and animation jobs will be disrupted by AI by 2026. Moonvalley intends to allow creators to request their content be removed from its models, let customers delete their data at any time, and offer an indemnity policy to protect its users from copyright challenges. Unlike some “unfiltered” video models that readily insert a person’s likeness into clips, Moonvalley is also committing to building guardrails around its tools. Like OpenAI’s Sora, Moonvalley’s models will block certain content, like NSFW phrases, and won’t allow users to prompt them to generate videos of specific people or celebrities. “We founded Moonvalley to make generative video technology that works for filmmakers and creative professionals,” Moonvalley wrote in a blog post in March. “That means addressing fear and distrust, as well as solving technical problems that keep generative AI from being a realistic tool for professional production.”

TechCrunch is joining forces with VivaTech, Europe’s biggest startup and tech event, to select startups for the prestigious VivaTech Innovation of the Year at VivaTech 2025. This partnership will highlight exceptional creativity, technological ingenuity, and industry-transforming potential among exhibiting startups. The VivaTech Innovation of the Year Award acknowledges and celebrates exhibiting startups at VivaTech 2025 that showcase outstanding innovation, advanced technological capabilities, and significant potential to revolutionize their respective industry. TechCrunch will curate a shortlist of the top 30 candidates, appoint a jury member to help select the final top 5, and participate on stage at VivaTech on June 11 alongside the full jury during the final award presentation. On June 12, TechCrunch will also serve as the official host of the VivaTech Global Awards Ceremony, where the winners of the 5 VivaTech Awards will be announced, providing the Master of Ceremony to lead the event with introductions, award transitions, and the closing moment. Five finalists will receive a valuable opportunity to pitch on the VivaTech stage on June 11, gain introductions to venture capitalists, enjoy exclusive networking with jury members, achieve high visibility across VivaTech marketing channels, and receive two additional 4-day passes to attend VivaTech 2025.The Innovation of the Year award winner will also secure a place in the highly competitive Disrupt Startup Battlefield 200, with eligibility to present at TechCrunch’s flagship annual event held in San Francisco, CA in October 2025. VivaTech Innovation of the Year Award eligibility guidelines: Must be an exhibiting startup at VivaTech 2025 Founded in or after 2020 Fewer than 50 employees Selection of 5 Finalists – June 4 Finalists’ Pitch Session at VivaTech – June 11 VivaTech Global Awards Ceremony – June 12 Book your pass now About Viva Technology VivaTech accelerates innovation by connecting startups, tech leaders, major companies and investors responding to our world’s biggest challenges. Techcrunch event Join us at TechCrunch Sessions: AI Secure your spot for our leading AI industry event with speakers from OpenAI, Anthropic, and Cohere. For a limited time, tickets are just $292 for an entire day of expert talks, workshops, and potent networking. 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 REGISTER NOW Each year, over four exciting days in Paris, VivaTech creates Europe’s biggest technology and startup event, exploring the most disruptive topics in tech with world-premiere demos, launches and conferences in a collaborative ecosystem. This is where business meets innovation. Take part to the ninth edition of VivaTech 11-14 June 2025.

Swiss tech startup Metafuels has unveiled plans to open its first commercial-scale sustainable aviation fuel (SAF) plant in the Port of Rotterdam.  Metafuels’ CEO Saurabh Kapoor told TNW that Turbe represents a “major step forward” toward ramping up SAF production. The startup also announced plans to build a similar facility in Denmark last year. “Europe has ambitious decarbonisation targets, but without scalable and affordable SAF production, aviation will struggle to keep up,” said Kapoor.  The facility, dubbed Turbe, will be built in collaboration with liquid energy storage provider Evos. Turbe will be integrated into Evos’ existing Rotterdam terminal, which offers access to the infrastructure needed to store green methanol in large quantities.   Register now Metafuels’ “aerobrew” technology converts renewable methanol into jet fuel, using a process it claims delivers high energy efficiency and up to 90% lower life cycle emissions than conventional jet fuel. The resulting SAF is “drop-in ready,” requiring no changes to aircraft or airport infrastructure. Turbe will be able to process both bio-methanol — sourced from biological waste — and e-methanol, which is made using renewable electricity and captured CO2. This flexibility allows Metafuels to respond to shifts in feedstock availability and regulatory demand, the company said. Kapoor said he expects both the Netherlands and Denmark plants to start producing fuel for use in commercial aircraft from 2028. However, before that happens, the company will need to obtain accreditation for its aerobrew process, which it expects to get by the end of this year.   Metafuels says the new site will produce 12,000 litres of SAF per day during its first phase, with ambitions to scale that tenfold in the second phase. For reference, a Boeing 737 Max has a fuel tank capacity of around 26,000 litres, which equates to a range of around 6,570 km. Metafuels and the broader aviation industry still has a long way to go to meet global and regional targets for SAF adoption.  SAF made up only 0.53% of global aircraft fuel use in 2024. The EU wants to ramp that up to 70% by 2050, which will require a drastic increase in the production of the greener fuel.   Story by Siôn Geschwindt Siôn is a freelance science and technology reporter, specialising in climate and energy. From nuclear fusion breakthroughs to electric vehic (show all) Siôn is a freelance science and technology reporter, specialising in climate and energy. From nuclear fusion breakthroughs to electric vehicles, he's happiest sourcing a scoop, investigating the impact of emerging technologies, and even putting them to the test. He has five years of journalism experience and holds a dual degree in media and environmental science from the University of Cape Town, South Africa. When he's not writing, you can probably find Siôn out hiking, surfing, playing the drums or catering to his moderate caffeine addiction. You can contact him at: sion.geschwindt [at] protonmail [dot] com Get the TNW newsletter Get the most important tech news in your inbox each week.