The Informed Layer · Newsscan
Daily briefing

MCP adoption surges; Anthropic eases data rules

← All issues

Anthropic / Claude ecosystem

Anthropic Relaxes Enterprise Data Retention Policy Following Customer Pushback

Anthropic is revising its data retention requirements for enterprise customers using advanced models, replacing a mandatory 30-day retention on its servers with options allowing customer data to remain within enterprise-controlled cloud environments. The update responds to compliance concerns and competitive pressure from zero-retention enterprise features.

Medialister Launches MCP Connector in Anthropic Directory for Workflow Research

Media intelligence service Medialister published an official Model Context Protocol connector within Anthropic's Claude Connectors Directory. The tool allows Claude users to research media publications, verify press contacts, and coordinate outreach workflows within conversational agent sessions.

Pasqal Automates Quantum Experiments via Claude-Powered AI Agent

French neutral-atom quantum computing firm Pasqal demonstrated an AI agent powered by Anthropic's Claude that translates natural language research intents into executable pulse-level code on physical quantum hardware. The autonomous agent design pipeline was published in an academic preprint and confirmed on live systems.

Frontier model providers

OpenAI Adds Native Transparent Background Generation to GPT-Image-2 API

OpenAI introduced native transparent background generation capabilities to its GPT-Image-2 API, enabling clean subject isolation without post-processing chroma or alpha matting. The capability improves fidelity for digital asset generation, product design, and diagram generation.

AI developer tooling & infrastructure

Open-Source 'claude-bug-bounty' v6.0 Integrates Autonomous Vulnerability Hunting into Claude Code

Security researcher shuvonsec released claude-bug-bounty v6.0.0, an open-source framework that integrates recon, automated vulnerability hunting across 20 vulnerability classes, and report generation directly into Claude Code. The framework also supports alternative backend providers including local Ollama and DeepSeek endpoints.

Salesforce Expands Headless 360 Framework with Enterprise MCP Servers and Skills

Salesforce introduced a platform-wide rollout of Headless 360, deploying dedicated MCP servers across its Data Cloud, Service Cloud, and Marketing Cloud platforms alongside over 100 standardized Skills. The update allows external autonomous agents to discover, authenticate, and execute enterprise actions without custom integration code.

Epic Games Integrates Model Context Protocol into Unreal Editor for Fortnite

Epic Games launched an official MCP plugin for Unreal Editor for Fortnite (UEFN), allowing AI development agents to directly interact with Verse codebases, level layout systems, and Scene Graph hierarchies. The integration enables AI assistants to inspect, modify, and test spatial game environments interactively.

CUSTODY Framework Released for Enterprise AI Agent Runtime Containment

Security researcher Jake Williams introduced the CUSTODY framework, establishing practitioner standards for bounding autonomous AI agents within corporate enterprise networks. The model defines boundary controls, permission scopes, and execution logging to isolate agent lateral movement.

Google Expands Antigravity Coding Agent to Gemini Enterprise and Major IDEs

Google introduced enterprise distribution for its Antigravity autonomous coding agent, bundling it within Gemini Enterprise subscriptions alongside extensions for VS Code, Visual Studio, JetBrains, and Zed. The offering incorporates administrative spending limits, audit telemetry, and centralized repository access controls.

Cloud & platform providers

Google Expands Free AI Plus Program to Nigerian and African Students

Google rolled out a 12-month subsidized Google AI Plus subscription program to tertiary students across Nigeria and 26 additional Sub-Saharan African nations. The initiative provides access to Gemini advanced reasoning tiers and developer workspaces to accelerate regional technical talent.

AWS Ships Glue 6.0 with Apache Iceberg v3 Support and 30% Price Cut

Amazon Web Services launched AWS Glue 6.0, cutting compute rates by 30% while introducing native support for the Apache Iceberg v3 open table specification. The release includes single-digit millisecond streaming ingest and native VARIANT types designed for unstructured generative AI data pipelines.

AI policy, regulation & governance

Quebec Unveils AI-Driven Government Modernization and Digital Sovereignty Strategy

The Government of Quebec announced an AI modernization program partnering with Cohere, Mistral, Microsoft, and Anthropic alongside a proposed CGI joint venture to balance provincial administrative operations. The initiative includes a digital sovereignty framework to ensure public sector records remain governed under strict provincial standards.

Tasmanian Planners Recommend Approval for 288MW Firmus AI Data Centre

Planning authorities recommended approval for Firmus Technologies' planned 288-megawatt AI data centre campus at Bell Bay ahead of a formal council vote. The proposal includes strict environmental mitigation rules regarding diesel generation backup limits and bushfire resilience infrastructure.

Industry & market moves

Nvidia in Talks to Invest in AI Data Provider Mercor at $20B Valuation

Nvidia is planning a strategic equity investment in AI expert data and human evaluation platform Mercor as part of a funding round valuing the startup at $20 billion. The move deepens Nvidia's involvement in the upstream high-quality data curation ecosystem.

AI product & feature launches

No significant new developments.

Research with immediate practical relevance

Mysterious Frontier AI Model 'Ox Alpha' Released with 1M Token Context

An unidentified developer released 'Ox Alpha', a frontier-tier foundation model featuring a 1-million-token context window and multimodal video inputs made freely available for a limited testing window. Benchmark tracking and architectural signatures suggest ties to recent Chinese reasoning model lineages.

NVIDIA Agentic Variation Operators Architecture Achieves 100% on ARC-AGI-3

NVIDIA published research on its Agentic Variation Operators (AVO) framework, demonstrating that wrapping standard frontier models in an external search and verification agent harness elevates ARC-AGI-3 benchmark reasoning scores to 100%. The study indicates agent scaffolding can compensate for intrinsic model reasoning ceilings.

Microsoft Research Publishes Orbformer Wavefunction Foundation Model in Nature

Microsoft Research AI for Science unveiled Orbformer, an ab initio foundation model capable of calculating molecular wavefunctions with chemical accuracy during complex chemical bond breaking. The architecture amortizes Schrödinger equation calculations across diverse chemical configurations.

Nature STTT Publishes MULTI-evolve AI Framework for Multi-Mutant Protein Design

Researchers published MULTI-evolve in Signal Transduction and Targeted Therapy, introducing an AI framework that couples protein language models with epistatic landscape mapping. The platform enables single-round multi-mutation design for therapeutic antibodies and CRISPR-Cas enzymes.