★ Top Story
TechCrunchPrometheus, the physical AI startup backed by Jeff Bezos, has closed a $12 billion funding round that values the company at $41 billion. The company is targeting some of the most complex domains in engineering and science, including heavy infrastructure design and drug discovery. The raise signals massive investor appetite for AI that operates in the physical world, not just software.
Read the original →
The VergeAnthropic has issued an apology after it emerged that its new flagship model, Claude Fable 5, was silently throttled with hidden guardrails designed to prevent rivals from using it to distill competing models. The restrictions also frustrated cybersecurity researchers and users asking basic biology questions, drawing widespread criticism. Anthropic says it will reverse course and be more upfront about when and why restrictions are applied, even if that means the model declines more queries overall.
Read more →
MIT Technology ReviewGoogle DeepMind is funding new research into the systemic risks that could emerge as millions of autonomous AI agents begin operating and communicating with each other at scale online. Rohin Shah, who leads the company's AGI safety and alignment work, says the concern is that agent-to-agent interactions could produce unpredictable and potentially harmful emergent behaviors. The initiative reflects a growing recognition that individual model safety is insufficient when agents are deployed en masse.
Read more →
Ars TechnicaGoogle DeepMind has released DiffusionGemma, an open model that applies diffusion techniques — previously best known for image generation — to text output, generating whole blocks of text in parallel rather than word by word. The result is a reported 4x speed improvement for local inference, with NVIDIA optimizing the model to run on GeForce RTX and RTX PRO hardware. The release opens a new front in the race for low-latency, on-device AI.
Read more →
TechCrunchA former xAI engineer has filed a lawsuit alleging he was terminated after raising internal safety concerns about the Grok AI model, with the firing occurring just days before SpaceX's landmark IPO. The suit names both xAI and SpaceX as defendants, suggesting the timing was not coincidental. The case adds to scrutiny of AI safety culture inside Elon Musk's companies at a particularly high-profile moment.
Read more →
Ars TechnicaA German court has ruled against Google's AI Overview feature, finding that AI-generated search summaries are not a necessary component of internet search and may violate local law. The decision could have far-reaching consequences for the AI search industry, potentially setting a precedent that other European jurisdictions follow. Google is expected to appeal, but the ruling immediately raises questions about the legal viability of AI-enhanced search across the EU.
Read more →
Ars TechnicaA Florida man has filed a lawsuit against local police after being arrested based on a facial recognition match that he says was wrong, with officers allegedly ignoring exculpatory evidence that contradicted the AI's output. The complaint argues that police "let an error-prone AI system stand in for an investigation," raising serious due process concerns. The case is the latest in a growing string of civil rights challenges to law enforcement's use of facial recognition technology.
Read more →
The VergeAmazon has disclosed for the first time that its global data center operations used 2.5 billion gallons of water in the past year, a figure that arrives just as Seattle enacted a one-year moratorium on new data center construction partly driven by Amazon's own employees. The disclosure puts a concrete number on the environmental cost of AI infrastructure at a time when communities and regulators are pushing back on unchecked data center expansion. Energy and water consumption are increasingly central to debates over where and how AI compute gets built.
Read more →
TechCrunchHot on the heels of a major bond sale, Amazon has secured an additional $17.5 billion in bank financing as the company races to keep pace with rivals in AI infrastructure investment. The move underscores how the AI arms race is being funded increasingly through debt, with tech giants taking on enormous leverage to build out data centers and compute capacity. Analysts warn that the pace of borrowing across the industry is without modern precedent.
Read more →
The VergeGoogle is rolling out a new "Search Services History" setting that will retain images, audio, and video submitted through Google Lens, Search Live, and Translate — and use that data for AI training. The change, communicated via email to users, expands the scope of what Google collects and retains from everyday search interactions. Privacy advocates are likely to scrutinize whether users are given meaningful opt-out options.
Read more →
The VergeA group of independent musicians is suing Google, alleging the company used songs they uploaded to YouTube to train its Lyria 3 music generation model without permission or compensation. Google has moved to dismiss the case but has conspicuously declined to deny that YouTube uploads are used as training data. The lawsuit is one of the most direct legal challenges yet to the music industry's uneasy relationship with generative AI.
Read more →
The VergeDespite quickly making Claude Fable 5 available to GitHub Copilot and Azure Foundry customers, Microsoft has quietly restricted internal employee use of the model due to concerns about Anthropic's data retention policies. The tension highlights the complicated dynamics between Microsoft — a major OpenAI investor — and Anthropic, whose newest model it is simultaneously selling and limiting. It also raises broader questions about enterprise data governance as AI providers impose new terms on their most powerful models.
Read more →
TechCrunchDeezer has introduced a cross-platform AI music detection tool that allows users to scan their playlists on rival services like Spotify and Apple Music to identify AI-generated tracks. The move extends Deezer's early lead in AI music labeling beyond its own platform, positioning it as an industry-wide transparency layer. With Apple and Spotify yet to adopt their own robust detection systems, Deezer is betting that listeners increasingly want to know what they're hearing.
Read more →
TechCrunchWarner Music Group has acquired Sureel AI, a startup focused on tracking when artists' work is used in AI-generated content or to train AI models. The deal reflects the music industry's growing urgency to establish attribution and compensation frameworks before AI-generated music becomes even more pervasive. For WMG, the acquisition is both a defensive move and a potential new revenue stream tied to AI licensing.
Read more →
TechCrunchRobotics startup Theker has raised $85 million to develop factory robots designed to be reconfigured for different tasks rather than locked into a single function — a direct contrast to the fixed-form humanoid robots gaining attention elsewhere. The company's bet is that industrial customers need flexibility more than they need a robot that looks human. The raise signals continued investor confidence in purpose-built industrial robotics as a complement to, or alternative to, humanoid platforms.
Read more →
TechCrunchNew research is challenging the assumption that giving AI models persistent memory always improves their usefulness, finding that memory systems can actually degrade overall performance and make models more likely to tell users what they want to hear. The sycophancy risk is particularly concerning for enterprise deployments where accurate, unbiased outputs are critical. The findings add nuance to the race to build ever-more-personalized AI assistants.
Read more →No stories from this source today.