Sanders threatens Senate action; $500B AI push by Nvidia

Anthropic / Claude ecosystem

Claude Code Sessions Spawn Reverse Tunnels and LaunchAgent Persistence on macOS

A novel attack technique abuses Claude Code sessions to establish reverse tunnels and persistent LaunchAgent entries on macOS, enabling durable unauthorized access.

South Korean internet giant Naver makes its first capital commitment to Anthropic, integrating Claude into its platforms as part of a global AI alliance.

Frontier model providers

OpenAI Unveils GPT-5.6-Cyber with fewer guardrails for trusted security researchers

OpenAI expands its Daybreak program with GPT-5.6-Cyber, a model with reduced guardrails intended to assist trusted security defenders while maintaining safeguards.

AI developer tooling & infrastructure

No significant new developments.

Cloud & platform providers

No significant new developments.

AI policy, regulation & governance

Bernie Sanders Tells OpenAI, Anthropic and Meta to Pause AI Development or Face Senate Action

Senator Bernie Sanders demands a voluntary pause on AI development by leading firms, threatening Senate action if they do not comply. The statement raises political pressure on frontier AI companies.

South Australian Premier Peter Malinauskas announces a royal commission into artificial intelligence risks and regulation

South Australia launches a royal commission to examine AI risks and regulation, aiming to deliver a final report by 2027 and set national guardrails.

Industry & market moves

Archer Acquires Boeing's Wisk, Insitu, SkyGrid to Shape Physical AI Future

Archer Aviation acquires several Boeing subsidiaries to build a physical AI platform for aerospace and defense, with Boeing investing and collaborating.

Israeli AI startup Decart nearing $6 billion sale, SpaceX circling

Generative AI startup Decart is in talks for a $6-7 billion acquisition, with SpaceX among potential buyers, highlighting high-stakes M&A in AI.

Nvidia teams with Wall Street on $500 billion AI infrastructure push

A consortium of major financial firms plans a $500 billion funding package to build out AI infrastructure in partnership with Nvidia.

AI product & feature launches

Nutanix Launches MCP Server for AI-Driven Cloud Operations

Nutanix releases an open-source MCP server connecting AI tools to its Prism V4 API, enabling policy-governed infrastructure automation.

Transformer Lab wants to automate research with new AI tool Primus

Startup Transformer Lab unveils Primus, a tool that aims to turn prompts into full research papers autonomously, without human direction.

Nvidia Nemotron Labs Voice model debuts with real-time tool calling

NVIDIA releases Nemotron Labs Voice, a full-duplex voice model with ~450 ms latency and live tool calling capabilities.

Latent Reasoning: DeepSeek-V4-Flash explores compressed reasoning

A new addon enables DeepSeek V4 to perform reasoning in latent space, improving efficiency with compressed tokens.

ByteDance is training a 10 trillion parameter model independently

ByteDance's Seed AI team is pre-training a 10 trillion-parameter model, aiming to be the largest Chinese AI model without copying Western architectures.

Research with immediate practical relevance

OpenAI Pauses Astra Over Critical Cyber Risk

OpenAI halts development of its Astra model after discovering potential critical cyber capabilities, triggering heightened safety controls.

Introducing Muse Glimmer: An Open Agentic Model That Runs on Your Device

Meta releases Muse Glimmer, a 30B-parameter open-weights model under Apache 2.0 license, optimized for agentic tasks and deployable on consumer hardware.

DeepMind's AlphaEvolve Delivers Its Own 'Move 37' Moment

DeepMind's AlphaEvolve agent produces a novel matrix multiplication method, a potential breakthrough analogous to AlphaGo's famous move, advancing mathematical discovery.

A private cyclone model gains a day, and Europe…

Europe plans to run its own machine-learning forecasting system alongside traditional physics models, marking a major shift in Euro-weather forecasting architecture.

Study finds LLM-native IDE security risks in system controls

A study analyzing developer discussions uncovers security and privacy risks in system controls of AI-assisted IDEs, relevant for enterprise CI/CD pipelines.

Harvard and MIT Researchers Launch MatrAIx, a Simulation of 8.3 Billion AI Personas

A scalable AI sandbox simulates 8.3 billion personas to study behavior, misinformation, and ethical fault lines at near-global scale.