A federal judge has ruled in Anthropic's favour, ordering the Trump administration to reverse its designation of the company as a supply chain risk and to stop directing federal agencies to cut ties with it. The ruling, from Judge Rita F. Lin of the Northern District of California, is a significant win for Anthropic in what has been one of the most dramatic standoffs between a tech company and the US government in recent memory.

To recap, the dispute started when Anthropic pushed back on the Pentagon's request to allow its AI models to be used for any lawful purpose without restriction. Anthropic drew a line at two things specifically, autonomous weapons systems and mass domestic surveillance. The government did not accept those limits. Defense Secretary Pete Hegseth then designated Anthropic a supply chain risk, a label typically applied to foreign adversaries, and President Trump ordered federal agencies to phase out use of the company's products. Anthropic sued shortly after.

In court, Judge Lin was pointed in her assessment of the government's actions, describing the orders as something that looked like an attempt to cripple Anthropic rather than a legitimate national security measure. She ultimately ruled that the designations violated the company's free speech protections and ordered the administration to back down.

The White House had spent the weeks leading up to the ruling publicly attacking Anthropic, calling it a radical-left company and framing its refusal to comply as a threat to national security. Anthropic CEO Dario Amodei had described the Pentagon's actions as retaliatory and punitive. The court appears to have agreed with that reading.

Anthropic said in a statement that while the lawsuit was necessary to protect the company and its partners, its focus remains on working with the government to ensure Americans benefit from safe and reliable AI. The case is not fully closed, but Thursday's injunction is a major milestone and gives Anthropic breathing room while the broader legal battle continues.

OpenAI has announced it will discontinue Sora, the AI video generation app it launched in September last year. The company has not given a specific reason for the decision, saying only that it needs to make trade-offs on products with high computing costs. Details on timelines for shutting down the app and API, as well as how users can preserve their work, are expected to follow soon.

Sora was positioned as OpenAI's play in the short-form video space, built to let users generate AI videos from text prompts. The ambition was to tap into the same attention economy that drives TikTok, YouTube Shorts, and Instagram Reels. But the product ran into significant headwinds from advocacy groups, academics, and safety researchers who raised concerns about its potential to generate deepfakes, nonconsensual imagery, and a broader flood of low-quality AI-generated content.

The shutdown also brings an early end to a high-profile $1 billion deal with Disney, announced just over three months ago. Under that three-year agreement, Disney had committed to investing $1 billion in OpenAI and lending more than 200 of its iconic characters for use in short AI-generated videos through Sora. With the app being discontinued, the future of that partnership is now unclear.

It is a notable retreat for a company that has been on an aggressive expansion run. OpenAI closing a product so shortly after a major corporate partnership is the kind of move that raises questions about how the company is making decisions about where to focus its resources, especially as it continues to pour investment into ChatGPT, GPT-5.4, and its core API business.

While OpenAI just announced it is shutting down Sora, ByteDance is moving in the opposite direction. The company has confirmed that Dreamina Seedance 2.0, its new AI audio and video generation model, is now rolling out inside CapCut, its popular video editing app. The launch is limited for now, starting with users in Brazil, Indonesia, Malaysia, Mexico, the Philippines, Thailand, and Vietnam, with more markets expected to follow.

The limited rollout is not accidental. ByteDance had paused a wider global launch of Seedance 2.0 after its debut in China caused an immediate backlash from Hollywood studios over copyright concerns. Videos generated by the model went viral, some featuring well-known celebrities in scenes they never filmed, and major studios including Disney sent cease-and-desist letters. The fact that the model is now live in selected markets but still absent from the United States strongly suggests the legal and technical cleanup work is still ongoing.

What the model can actually do is genuinely impressive. Creators can generate video content from just a text prompt, a reference image, or an existing clip, with the model handling realistic textures, lighting, and movement across different angles and perspectives. It supports clips of up to 15 seconds across six aspect ratios and can be used for everything from cooking content and fitness tutorials to product demonstrations and action-heavy videos, a category where AI video has historically struggled.

ByteDance has added some safeguards this time around. The model will not generate video from images or clips containing real faces, and CapCut will block the generation of content that uses unauthorised intellectual property. Every piece of content produced will also carry an invisible watermark to help identify AI-generated material when it is shared off-platform.

Google has launched Lyria 3 Pro, an upgraded version of its AI music generation model, just a month after releasing the original Lyria 3. The biggest practical difference is track length. Where Lyria 3 could generate clips of up to 30 seconds, Lyria 3 Pro extends that to three full minutes, which moves it much closer to something you could actually use as finished music rather than just a sound snippet.

Beyond length, the Pro model offers more creative control over how a track is structured. Users can specify individual sections in their prompt, including intros, verses, choruses, and bridges, and the model understands how those pieces fit together in a way its predecessor did not. That kind of structural awareness is a meaningful step up for anyone trying to produce something that sounds like a real song rather than a randomly generated audio clip.

The model is rolling out in the Gemini app, though access there is limited to paid subscribers. It is also coming to Google Vids, the company's AI-powered video editing tool, and ProducerAI, the AI music production platform Google acquired last month. On the enterprise side, Lyria 3 Pro is available in Vertex AI, the Gemini API, and AI Studio.

Google has been clear about how the model was trained, saying it used data from partners and permissible content from YouTube and Google. The company says the model does not imitate specific artists, but if a user names an artist in a prompt, it will take broad inspiration from their style to shape the output. All tracks generated with Lyria 3 or Lyria 3 Pro are automatically marked with a SynthID watermark so listeners and platforms can identify AI-made music.

The launch comes at a moment when the music industry is actively trying to get ahead of AI-generated content. Spotify this week released tools letting artists review music being released under their name to prevent misattribution. Deezer has built detection tools it is now making available to other streaming platforms.

An AI-driven app designed to manage family life. It handles scheduling, task delegation, and meal planning, and allows users to add tasks via voice or photo