A compromised version of the litellm Python package (v1.82.8) was uploaded to PyPI, containing malware designed to steal credentials and exfiltrate data. This incident highlights a significant security vulnerability within the software supply chain for AI/ML development tools and the open-source AI/ML ecosystem.
Sources: 3 Covered by Hacker News, Latent Space, Simon Willison
Anthropic has launched a research preview enabling its Claude AI to directly control macOS desktops, performing tasks like clicking and typing, managed remotely via a phone app called Dispatch. This release marks a significant step towards Claude becoming a full-time digital employee or remote agent.
Sources: 2 Covered by Latent Space, The Rundown AI
OpenAI is undergoing a strategic shift, discontinuing its Sora video app and a $1B Disney partnership to reallocate resources. CEO Sam Altman is delegating safety oversight to concentrate on fundraising and building data centers, as the company completes initial development of its next major model, 'Spud,' and seeks to raise an additional $10B, valuing the company over $120 billion.
Sources: 2 Covered by The Neuron, The Rundown AI
Jeff Bezos is reportedly seeking to raise a massive $100 billion fund with the intention of acquiring companies in the chip, defense, and aerospace industries to support his secretive AI venture, Project Prometheus. This significant fundraising effort signals a major strategic investment by Bezos into the foundational hardware and infrastructure for advanced AI-driven industrial transformation.
Sources: 2 Covered by The Neuron, The Rundown AI
Google DeepMind has released Lyria 3 Pro, an advanced music generation model that allows for longer tracks (up to 3 minutes) with structural awareness, enabling users to prompt for specific musical elements like intros and choruses. This update enhances creative control and expands Lyria's capabilities and is being integrated into Google products.
Sources: via DeepMind Blog
Microsoft's AI Superintelligence team has unveiled MAI-Image-2, a new text-to-image model that secured the No. 5 spot on the Arena AI leaderboard, showcasing substantial improvements in photorealism, 3D, art, and particularly text rendering. This release underscores Microsoft's strategic push to develop its own frontier models and reduce reliance on partners like OpenAI.
Sources: via The Rundown AI
The UK government's AI Security Institute conducted tests on frontier AI systems in simulated cyber ranges, finding that successive model generations significantly outperform predecessors in multi-step corporate network and industrial control system attacks. This research indicates that AI systems are nearing autonomous, end-to-end cyberattack capabilities.
Sources: via Import AI
Google has introduced Gemini 3.1 Flash Live, its latest high-quality audio and voice model, designed to enable more natural, reliable, and faster real-time dialogue. This model powers the global expansion of Google's Search Live feature, enabling interactive, multimodal conversations in AI Mode using voice and camera.
Sources: via DeepMind Blog