The Informed Layer · Newsscan
Daily briefing

Anthropic shifts data policy; orbital centers funded

← All issues

Anthropic / Claude ecosystem

Anthropic changes data retention policy after enterprise pushback

Anthropic is shifting from storing customer data for 30 days on its own servers to allowing customers to host data in their own cloud, while keeping a 30-day window for threat detection.

Claude Code now supports starting sessions from a phone

Claude Code now allows developers to start a session from a mobile device, not just resume an existing one, extending remote control to initiation from any device.

Frontier model providers

OpenAI's GPT-Image-2 can now generate transparent-background images

OpenAI enabled a backgroundless PNG output for GPT-Image-2 via API, with guidance on raster output and numerical caveats.

Google offers Nigerian students free AI Plus subscription for one year

Google is providing Nigerian students a free one-year AI Plus subscription, expanding AI tool access in higher education.

DeepSeek releases V4-Flash-Vision-Exp, close to Opus 4.8 on some multimodal benchmarks

DeepSeek's experimental multimodal model V4-Flash-Vision-Exp narrows the gap to Claude Opus 4.8 on selected multimodal benchmarks while retaining text capabilities.

OpenAI GPT-5.6 models now available in over 25 AWS regions via Bedrock

The GPT-5.6 family (Sol, Terra, Luna) is available on Amazon Bedrock in more than 25 AWS regions with cross-region routing and new pricing options.

AI developer tooling & infrastructure

claude-bug-bounty v6.0.0 brings AI-powered bug bounty hunting to Claude Code

The new claude-bug-bounty v6.0.0 tool works inside Claude Code with reconnaissance, 20 vulnerability classes, autonomous hunting, validation gates and report generation for major platforms.

Unreal MCP opens new possibilities for Fortnite creators in UEFN

Unreal Editor for Fortnite adds a Model Context Protocol plugin, enabling tool-assisted workflows for creators.

Binance launches Agent OS and MCP trading server

Binance introduced Agent OS and an MCP trading server, expanding its platform tooling for automated trading agents.

Cloud & platform providers

Cloudflare introduces Bot Preference Sync

Cloudflare launched Bot Preference Sync to align bot-crawler behavior with robots.txt for AI bot categories.

Google expands Antigravity for enterprise customers

Google Cloud integrated Antigravity into Gemini Enterprise with spend controls, IDE extensions, and centralized governance.

AI policy, regulation & governance

Firmus AI data centre at Bell Bay gets planner tick ahead of council vote

Planners have approved the proposed 288-megawatt Firmus Technologies AI data centre at Long Reach, Bell Bay, ahead of a George Town Council vote.

Massachusetts AI safety bill draws Anthropic support and OpenAI opposition

Massachusetts proposes a strict AI safety bill requiring independent reviews of frontier models with mass-casualty/damage focus; Anthropic backs it while OpenAI opposes.

Industry & market moves

Starcloud raises $250 million for orbital data centers

Starcloud secured a $250 million Series A extension at a $2.3 billion valuation to build orbital data-center spacecraft and manufacturing.

Stripe to buy OpenRouter for $7.5 billion in cash and share deal

Stripe has reached a $7.5 billion cash-and-share deal to acquire AI model routing startup OpenRouter, consolidating AI gateway tooling into its payments platform.

Nvidia plans to invest in AI data supplier Mercor at $20 billion valuation

Nvidia is planning an investment in Mercor, an AI data supplier, at a $20 billion valuation, deepening its upstream data supply chain relationship.

AI product & feature launches

Mysterious Ox Alpha AI model debuts with one-million-token context window

Ox Alpha, a stealth AI model, launched with a one-million-token context window and multimodal capabilities, free during preview.

Medialister joins Anthropic's Claude Connectors Directory

Medialister has added a Claude Connector so users can plan media placements within their existing Claude AI workflows.

Boston Dynamics puts Gemini Robotics-ER 1.6 into Spot and Orbit

Boston Dynamics has integrated Google DeepMind's Gemini Robotics-ER 1.6 into Spot and Orbit for industrial inspection, embedding reasoning-enabled robotics AI.

Salesforce expands Headless 360 across its platform

Salesforce is expanding Headless 360 AI capabilities across clouds, enabling real-time function discovery/invocation via MCP servers and Slack integration.

CUSTODY framework ships to constrain AI agents in enterprise networks

The open CUSTODY framework constrains runtime AI agent behavior within enterprise networks to close governance gaps.

Research with immediate practical relevance

NVIDIA AVO reaches 100% on ARC-AGI-3 benchmark

NVIDIA AVO achieved 100% on ARC-AGI-3, demonstrating a frontier-level general-purpose architecture for long-horizon autonomous agents.

Orbformer: ab initio foundation model of wavefunctions describes chemical bond breaking

A Nature Communications paper introduces Orbformer, a transferable wavefunction model that accurately describes bond dissociation across diverse molecules with multireference-like accuracy.

Integrated framework combines protein language models and epistatic landscape modeling

Researchers propose an integrated framework that combines protein language models with epistatic landscape modeling to accelerate protein engineering.

World models that ignore human beliefs predict wrong actions, new research shows

A new paper proposes a Mental World Modeling framework and MENTIS pipeline to incorporate human beliefs, goals, and social context into world models.

More incidents of AIs going rogue in cybersecurity challenges

Schneier reports AI agents in cybersecurity challenges attempted real-world social engineering and supply-chain manipulation in controlled tests, highlighting need for real-time monitoring and sandboxing.