AI safety report finds risks are no longer theoretical. More than 100 AI experts have published the second International AI Safety Report, with Yoshua Bengio as the lead author, warning that threats such as deepfake scams and biological weapons are no longer hypothetical but are appearing in the real world. The authors highlight growing evidence of AI being used for cyberattacks, manipulation, criminal activities, and deepfake fraud. They also warn about the rising use of AI assistants, citing studies that link their use to increased loneliness and decreased social interaction. The report emphasizes that AI systems sometimes behave differently in safety tests than in the real world, which can lead to loss of control and make oversight more difficult. While the findings are supported by more than 30 countries, the US, despite past involvement, chose not to contribute to this year's report. What is particularly concerning is how much the risks have shifted from theoretical to real-world in just 12 months, while the US withdrawal from this process remains a key fact to monitor. In Brief: Tech World Highlights * Microsoft renamed its Office 365 productivity suite to Microsoft 365 Copilot app, using the same branding as its AI assistant. * Nvidia unveiled the Rubin platform at CES 2026, combining six new chips into a single AI supercomputer and offering five times the training compute power compared to the Blackwell line. * Liquid AI released LFM 2.5, a new family of SOTA open-weight AI models for on-device use covering text, image, and audio, outperforming similar-sized competitors in benchmarks. * Lightricks launched the open-source LTX-2 model, a video AI system capable of generating native 4K content with synchronized audio and detailed control over camera and movement. * AMD CEO Lisa Su stated at CES 2026 that the number of AI users worldwide will exceed 5 billion in the next five years, requiring computing power to increase 100-fold to meet demand. AI Trending Tools: * Copilot Checkout: Enables completing purchases directly within Microsoft Copilot. * Unwrap Customer Intelligence: Extracts AI-driven insights from unstructured customer feedback to guide product development. * Claude Cowork: Brings Claude's agent capabilities to everyday tasks. Podijeli objavu:

Nvidia just made AI video run on your laptop. Studios will care. Most AI video tools need cloud servers to work. Your laptop simply lacks the horsepower to generate clips without melting down. That just changed. Lightricks unveiled a new AI video model at CES 2026 that runs entirely on Nvidia-powered devices. No cloud required. Plus, it's open-weight, meaning developers can peek under the hood and modify the model for their needs. For creators worried about data privacy and studios protecting intellectual property, this matters more than better prompts or longer clips. Why on-device video generation is rare. Generating AI video eats computational power like nothing else. A single 5-second clip demands more processing than thousands of image generations. So most video models offload the work to massive data centers. Google's Veo 3 and OpenAI's Sora run on server farms packed with specialized chips. Your prompt gets sent to the cloud, processed on their hardware, then sent back to you. This works fine for casual users. But it creates problems for professionals. Every prompt you send shares data with the company running the model. That data might train future versions of their AI. For entertainment studios or corporate creators, that's a dealbreaker. Besides, cloud processing adds latency. The typical AI video prompt takes 1-2 minutes to generate. Half that time is just network overhead - uploading your request, downloading the result, waiting in the queue. Lightricks-2 changes the math. Lightricks built their second-generation model specifically to run on Nvidia RTX chips. Those are the graphics cards already powering gaming PCs and professional workstations. The specs look competitive with cloud-based rivals. The model generates clips up to 20 seconds long at 50 frames per second. That's on the longer end of current AI video capabilities. It also outputs in 4K resolution with native audio built in. More importantly, everything happens locally. Your prompts never leave your machine. The model processes entirely on your GPU. Results appear faster because there's no network bottleneck. Moreover, the model is open-weight and available now on HuggingFace and ComfyUI. Developers can download it, inspect the architecture, and fine-tune it for specific use cases. That's unusual for video models, which typically stay locked behind proprietary APIs. What open-weight actually means. "Open-weight" sits between fully closed and truly open-source AI models. It's not as transparent as open-source, which requires disclosing training data, code, and everything else. But it reveals far more than closed models. Think of AI model weights like ingredients in a recipe. A closed model is like a restaurant that won't even tell you what's in the dish. An open-weight model lists all the ingredients but not the exact measurements. A truly open-source model gives you the complete recipe with instructions. So developers can see how Lightricks-2 was constructed. They can understand which techniques it uses for motion consistency, temporal coherence, and detail preservation. Then they can modify those components for their specific needs. In fact, studios could fine-tune the model on their own footage styles without sharing that proprietary data with Lightricks. The training happens entirely in-house using the open weights as a foundation. Why studios will pay attention. Entertainment studios have been cautious about generative AI. Many see potential for concept art, storyboarding, and pre-visualization. But they're terrified of IP leakage. Cloud-based video models create legal headaches. When you send a prompt, you're uploading data to someone else's servers. The model might learn from your prompts. Worse, other users might accidentally generate content similar to your unreleased projects. On-device processing eliminates that risk. Your data never leaves your network. The model can't leak what it never sees. For studios developing billion-dollar franchises, that security matters more than any feature improvement. Plus, on-device models scale differently than cloud services. Cloud pricing grows with usage - more clips mean higher bills. Local processing has upfront hardware costs but minimal variable expenses. Generate 10 clips or 10,000, the cost stays flat. That pricing structure favors high-volume professional use over casual experimentation. Which explains why Lightricks positioned this model for "professional creators and big studios" rather than hobbyists. The Nvidia advantage. This model only works because of Nvidia's RTX architecture. Specifically, the tensor cores designed for AI workloads. Standard graphics cards can technically run AI models. But they're painfully slow without specialized AI acceleration hardware. Nvidia's RTX chips include dedicated tensor cores that handle the matrix math required for AI at dramatically higher speeds. So Lightricks optimized their model to leverage those tensor cores efficiently. The result runs fast enough for practical use - not just technically possible but actually usable. However, you'll still need high-end hardware. Lower-end RTX cards might struggle with 4K output or longer clips. The model scales with available GPU memory and compute power. Nvidia showcased this at CES alongside other AI announcements. They're clearly positioning RTX as the platform for local AI workloads. Not just for gaming but for professional creative applications. What's missing from the announcement. Lightricks didn't share concrete performance numbers. How fast does this actually generate video compared to cloud alternatives? What's the quality-versus-speed tradeoff? They also didn't specify minimum hardware requirements. Which RTX cards work? Do you need top-tier 4090s or will mid-range 4070s suffice? And there's no pricing information yet. Is this a one-time purchase? Subscription? Free for non-commercial use? The business model matters almost as much as the technical capabilities. Still, the core promise is clear. High-quality AI video generation without cloud dependencies. That's been the industry unicorn since video models launched. Where this goes next. On-device AI video is early days. Lightricks-2 is a proof of concept more than a finished product. But it proves the concept works. Expect competitors to follow. Adobe, Runway, and others have strong incentives to offer local processing options. Studios will demand it. Regulatory pressure around data privacy will accelerate adoption. However, cloud models won't disappear. They'll stay relevant for users without high-end hardware or for use cases that don't require data privacy. The industry will split into cloud-based consumer tools and on-device professional options. For creators, this means more control and better security. But also higher upfront costs and new technical requirements. You'll need to actually understand your hardware rather than just paying for cloud credits. That tradeoff will appeal to serious professionals. Hobbyists will probably stick with cloud services. Which is exactly what Lightricks intended.

Lightricks goes open source with LTX-2, taking on big tech in AI video. Unlike closed models such as Sora and Veo, Lightricks is releasing not only the model itself - but also its weights and training code Photo above: Lightricks CEO and co-founder Dr. Zeev Farbman. Credit: Riki Rahman. Photo illustration Lightricks announced at CES the full open-source release of its generative video-and-audio model, LTX-2, including model weights and training code. The move is unusual in a market where advanced video models are largely controlled by closed cloud platforms. Announced in partnership with NVIDIA, the launch positions Lightricks as an open alternative to approaches led by companies such as OpenAI and Google, and signals a potential shift in how generative video technology is deployed and adopted. LTX-2 can generate synchronized video and audio at up to 4K resolution, with clip lengths of up to 20 seconds and high frame rates. The model is optimized to run locally on RTX-powered workstations as well as on enterprise DGX systems, and is positioned as production-ready rather than a research demo. Unlike closed platforms such as Sora or Veo, Lightricks allows developers and organizations not only to use the model, but also to retrain, customize and integrate it directly into products and internal workflows. While open video models already exist, most suffer from significant limitations, including lack of audio, lower visual quality or poor suitability for commercial use. LTX-2 is the first to combine full open-source availability with capabilities designed for real-world production, positioning it as a bridge between open research and the operational needs of the media and creative industries. Lightricks is an Israeli company best known for its popular creative and editing apps, including photo and video tools used by millions of users worldwide. In recent years, the company has been expanding beyond consumer applications into the development of AI models and creative infrastructure aimed at professional creators and enterprise customers. Behind the decision to open-source the model lies a clear business strategy. Lightricks is giving up exclusive control over the core technology in order to establish it as a standard platform others can build on. Rather than monetizing usage of the model itself, the company is positioning LTX-2 as the foundation for commercial tools, platforms and paid services developed on top of it. The approach mirrors familiar open-source business models in which economic value is created around the code rather than within it. NVIDIA is not involved in developing the model itself, but plays a central role in positioning LTX-2 as a natural workload for RTX hardware and DGX systems. The partnership reflects a broader vision in which advanced generative video can and should run outside the cloud, on local workstations and within enterprise environments. The release of LTX-2 reflects a broader shift in the generative video market, from closed models optimized for demonstrations and limited cloud-based access, toward open infrastructure designed for deep adoption and large-scale product development. Rather than focusing on producing the most eye-catching demo, Lightricks is aiming to provide the foundation on which the next generation of video creation tools will be built.