Meta patches zero-day exploit in Muse macOS app that allowed agent control
Meta has released a hotfix for its Muse macOS app after a researcher discovered a vulnerability that could allow attackers to hijack the AI agent and manipulate user data.
Latest
JetBrains Launches JetBrains Air: An Open System for Agentic Software Development
JetBrains has introduced JetBrains Air, an open system designed to coordinate, execute, and govern agentic development across various IDEs, teams, and organizations.
Verda raises $189M in Series B to expand full-stack AI cloud capacity
Finnish AI cloud provider Verda has raised $189 million in a Series B funding round led by Emergence Capital, bringing its total funding to over $450 million.
AI in Education: Generative AI Shows Dual Impact on Critical Thinking and Problem-Solving
A systematic review of 173 international studies finds that generative AI acts as both a catalyst and a barrier to students' critical thinking, depending on whether it is used to challenge perspectives or as a shortcut to answers.
OneCLI Released as Open-Source Platform for Managing AI Agents in Teams
OneCLI provides a sandboxed environment and centralized management for AI agents, allowing teams to securely share credentials and access policies.
macOS 27 Update Removes Ability to Disable Apple Intelligence, User Reports
A user claims that following an upgrade to macOS 27, the toggle previously used to opt out of Apple Intelligence has been removed, forcing the feature to remain active.
Can gzip function as a language model? A new implementation uses beam search for text generation
A developer has released GziPT, a tool that uses the gzip algorithm to generate text by predicting byte sequences that compress most efficiently, demonstrating the equivalence between compression and prediction.
vLLM v0.30.0 Release — Expanded Model Support and Model Runner V2 Enhancements
vLLM v0.30.0 introduces support for DeepSeek-V4.1-Flash, GLM-5.3-Flash, and more, alongside performance optimizations for Model Runner V2 and new features like Fast Start and Watermarking.
Xiaomi Releases MiMo-V2.6-Pro, a High-Performance Multimodal Reasoning Model
Xiaomi has launched MiMo-V2.6-Pro, an open-weights Mixture of Experts (MoE) model featuring 1 trillion total parameters and a 1M token context window.
OpenAI Partners with Independent Mathematical Advisory Group AGMAI
OpenAI has announced a partnership with the Advisory Group on Mathematics and Artificial Intelligence (AGMAI) to seek guidance on the presentation and evaluation of AI-driven mathematical breakthroughs.
OpenAI Calls for US-Led International Standards for Frontier AI and Recursive Self-Improvement
OpenAI proposed that the United States should lead the development of global technical standards for frontier AI to manage risks like recursive self-improvement (RSI) and ensure human control.
Robin Williams' Daughter Slams "Terrible" AI Videos of Late Actor
Zelda Williams calls on fans to stop sharing AI-generated videos of her late father, Robin Williams, describing recent deepfake content as "robotic and terrible."
Elevated error rates for multiple Claude models
The Claude status page indicates elevated error rates for multiple models.
Cheaper AI Models Can Increase Coding Task Costs Due to Verbose and Fabricated Responses
Fiddler research reveals that Anthropic's Claude Haiku 4.5 cost significantly more per successful coding task than Claude Sonnet 5, despite having a much lower token price, due to excessive output lengths and fabricated execution results.
Amazon Blocks Proxy Purchasing by Meta's AI Agent 'Muse'; Shopify to Support Across All Stores
Amazon has reportedly blocked proxy purchasing on Amazon.com by Meta's personal AI agent 'Muse.' Meanwhile, e-commerce platform Shopify has announced plans to enable Muse payments across all its stores.
Jun Kim joins Hugging Face to support MLX and oMLX ecosystem
Jun Kim, the creator and maintainer of oMLX, has joined Hugging Face to accelerate development of MLX-based tools and strengthen the local AI ecosystem.
Transformers Adds Support for Efficient GGUF Model Execution
Hugging Face has integrated support for GGUF models into the Transformers library, enabling efficient local inference using ggml kernels, starting with Apple Silicon.
OpenAI Proposes Priorities and Principles for Independent Third-Party AI Safety Assessments
OpenAI has outlined four priority areas and core principles for effective third-party assessments to enhance accountability and safety in frontier AI development.
SpaceXAI Unveils Grok 4.7 with Enhanced Coding and Knowledge Work Performance
SpaceXAI has released Grok 4.7, claiming its most capable model for coding and knowledge tasks, offering improved reasoning and higher speed at a competitive price.
The Rise of "Spymarks": Hidden Signals That Trace Media Back to Users
Beyond traditional watermarking, new "spymarking" technologies in audio, images, and text enable covert tracking of users by embedding identifiers that survive metadata removal.