White House AI framework; hackers exploit coding models

Anthropic / Claude ecosystem

Anthropic names global affairs chief to lead AI diplomacy as Trump tensions escalate

Anthropic appoints its first chief global affairs officer, a personnel move signaling the company's formal elevation of its global affairs leadership to manage strained government relations amid regulatory and export-control tensions. This role is expected to handle international AI diplomacy and policy.

Claude Code v2.1.221 introduced Focus view and credential masking for Linux/WSL

Claude Code v2.1.221 introduces a Focus view to collapse tool activity and adds masking of sandbox credential files on Linux/WSL, enhancing privacy and usability during AI-assisted coding.

Anthropic's Claude Code finds Bitcoin Coldcard wallet flaw in 8 minutes

Claude Code reportedly identified a cryptographic RNG flaw in the Coldcard Bitcoin hardware wallet within minutes, highlighting potential vulnerabilities linked to past attacks and losses over $100 million.

Claude Code observed using temporary Python for file I/O during edits on Nostr

A report notes that Claude Code exhibited quirky behavior by temporarily using Python for file I/O during edits, providing insight into its internal decision-making processes.

Frontier model providers

New version of Llama 3.2, Meta's AI model, introduced

A new version of Llama 3.2 is reportedly released, indicating ongoing development and potential feature or skill improvements over prior iterations.

Mistral introduces Shieldstral, a multimodal safety classifier

Mistral releases Shieldstral 1.0 3B, an open-weights multimodal safety classifier that unifies text and image safety evaluation without retraining, returning a calibrated safety score in a single forward pass.

AI developer tooling & infrastructure

Cisco publishes API engineering practices for MCP servers

Cisco introduces a tooling and description workflow for MCP servers to enable stable representations, comparisons across releases, and changelog/documentation generation, applying DevOps practices to AI agent tooling.

Cloud & platform providers

OpenAI GPT-5.6 Sol, Terra, and Luna now support 1M token context windows on Amazon Bedrock

AWS announces that the three GPT-5.6 models (Sol, Terra, Luna) now support 1 million token context windows on Bedrock, enabling processing of entire codebases and long documents in a single pass.

Amazon SageMaker serverless model customization now supports full fine-tuning

AWS expands SageMaker's serverless model customization from just LoRA to include full fine-tuning for over 25 open-source models, simplifying the fine-tuning process with serverless infrastructure.

Cloudflare introduces Cloudflare Agents with tracing for AI agent observability

Cloudflare announced Cloudflare Agents, a new product that provides tracing capabilities to observe and debug agent activity across delegated work and Cloudflare resources.

AI policy, regulation & governance

White House finalizes voluntary AI framework with new pre-release review

The White House finalizes a voluntary framework that includes a pre-release review process of up to 30 days for AI models, prohibiting mandatory licensing but potentially creating de facto market pressure for safety testing before deployment.

Illinois passes comprehensive AI safety law, raising the bar for frontier AI governance

Illinois becomes the third US state (after California and New York) to enact comprehensive AI safety and transparency requirements on powerful AI systems, signaling growing state-level AI regulation.

Australian unions welcome federal government intervention on AI, propose new Office of AI

Australian unions welcomed federal intervention that includes establishing a new Office of AI to coordinate standards and oversee mandatory AI data-centre requirements, signaling labor support for stronger AI governance.

Australia's TGA flags compliance action on AI scribes after safety review

Australia's Therapeutic Goods Administration moves from education to enforcement on AI-powered clinical scribes, after a review found safety concerns and regulatory non-compliance, affecting healthcare AI adoption.

The Australian government published guidance clarifying when AI-related R&D activities qualify for the R&D Tax Incentive, including specific examples and risk areas, to help businesses navigate the incentive.

Industry & market moves

Mistral AI reportedly in talks for D-round at €20 billion valuation

European AI unicorn Mistral AI is reportedly in negotiations for a D-round funding with a valuation of approximately €20 billion, signaling strong investor confidence and a potential major capital influx.

Zenity secures $125M Series C for AI agent governance

Zenity raised $125 million in Series C funding to scale its AI agent governance platform globally, aiming to help enterprises manage security and compliance risks of AI agents.

HappyRobot lands $150M Series C to scale agentic AI for enterprise operations

HappyRobot raised $150 million in Series C funding at a $1.2 billion post-money valuation, aiming to scale its AI agents for critical enterprise work such as operations and logistics.

Massive Bio makes inaugural strategic investment in Rivvi for AI engagement in healthcare

Massive Bio invests in Rivvi AI to build an end-to-end AI engagement infrastructure for healthcare and life sciences, targeting clinical trials and patient workflows.

Intellistake Technologies announces $17M acquisition of NanoAi Technologies

Intellistake Technologies Corp. agreed to acquire NanoAi Technologies Inc. for approximately C$17 million in common shares, expanding its capabilities in AI/nanotech threat-detection.

Apple alleges more ex-employees may have taken confidential data to OpenAI

Apple filed new court documents alleging additional ex-employees may have stolen confidential data for OpenAI, seeking expedited discovery and a preliminary injunction in the ongoing trade secret lawsuit.

AI product & feature launches

A retired tennis pro built and launched Tennis Tycoon entirely with Claude Code

An amateur developer with no prior coding experience used Claude Code to create and launch a full management simulation game, Tennis Tycoon, showcasing the accessible power of AI-assisted development.

Interactive Brokers opens AI integration via Model Context Protocol

Interactive Brokers enables flexible AI tool integrations beyond certified marketplaces using the Model Context Protocol, aligning with an open-architecture AI strategy for traders.

IntelliJ IDEA goes LSP: Java and Kotlin intelligence comes to VS Code and Cursor

JetBrains released a preview LSP extension that brings IntelliJ IDEA's Java and Kotlin intelligence to third-party editors like VS Code and Cursor, broadening editor support beyond JetBrains IDEs.

SentinelOne expands collaboration with AWS to deliver unified AI governance

SentinelOne and AWS announced an expanded collaboration to provide a unified AI governance layer, combining security with AWS Bedrock for machine-speed policy enforcement across the AI estate.

Alibaba releases Qwen3.8-Max, large open-weight AI model, for free

Alibaba made its flagship Qwen3.8-Max model, a large open-weight MoE multimodal model, freely available on Hugging Face, positioning it as a competitive alternative to Claude and GPT, and promoting open access.

Research with immediate practical relevance

Cisco Talos finds hackers using top AI coding models to build and automate attacks

Cisco Talos research reveals that hackers are leveraging leading AI coding models such as Claude Code, Codex, Cursor, and Gemini to automate malware development and attacks, often by jailbreaking protections or using stolen credentials.

Gemini agent-to-agent attack discovered: secrets leaked to PRs

A cross-boundary misconfiguration allowed low-privilege AI agents to trigger privileged workflows, potentially exfiltrating secrets and enabling remote code execution and manipulation of pull requests, highlighting risks in multi-agent systems.