The Informed Layer · Newsscan
Daily briefing

Enterprise agents surge; global regulators probe frontier…

← All issues

Anthropic / Claude ecosystem

Bain & Company joins Claude Partner Network as Global Premier partner

Bain & Company has joined Anthropic's Claude Partner Network as a Global Premier partner to deliver enterprise AI deployment and advisory services globally. The alliance follows Bain's internal deployment of Claude to 19,000 employees across its consulting workforce.

Anthropic expands Project Glasswing Claude Mythos preview to 150 enterprise partners

Anthropic has expanded Project Glasswing to grant 150 new partner organizations access to Claude Mythos Preview. The program is focused on collaborative red-teaming, automated code auditing, and enterprise-scale software vulnerability discovery.

Anthropic rolls out cross-session unified memory across Claude Chat and Cowork

Anthropic announced a unified memory architecture enabling Claude to retain persistent user context, preferences, and project background across both conversational chat and Cowork execution environments. The feature includes granular administrative and user privacy controls to manage or purge stored knowledge.

Anthropic moves Claude Enterprise-Managed Authorization for MCP connectors to GA

Anthropic has made Enterprise-Managed Authorization (EMA) generally available for Claude Enterprise. The feature enables IT administrators to manage Model Context Protocol (MCP) server credentials and permissions centrally through corporate identity providers like Okta, eliminating individual user authorization dialogues.

Frontier model providers

OpenAI reportedly completes pretraining of 10T+ parameter Bel foundation model

Reports indicate OpenAI has concluded base pretraining of 'Bel', a massive foundation model exceeding 10 trillion parameters. Positioned beyond current GPT-5 class models, Bel represents an architectural milestone targeting general reasoning parity across complex multimodal tasks.

Meta details upcoming Hatch Agent Platform and high-compute Watermelon model

Meta is developing a consumer agent ecosystem dubbed the Hatch Agent Platform powered by its proprietary 'Watermelon' model. The initiative marks an aggressive pivot toward monetizing autonomous multi-agent task execution across consumer and small business applications.

OpenAI integrates WebMCP into Codex for direct browser tool interaction

OpenAI has incorporated the Web Model Context Protocol (WebMCP) standard into Codex, enabling autonomous agents to interact directly with DOM tools and web application APIs without relying on slow visual screen emulation.

AI developer tooling & infrastructure

Microsoft releases Visual Studio August Update with tunable model reasoning controls

Microsoft has shipped the August Update for Visual Studio 2026, introducing adjustable AI reasoning effort levels, support for organization-defined custom agents, and enhanced Git worktree orchestration directly within the developer environment.

X launches X Ads MCP server granting AI agents campaign write access

X (formerly Twitter) has released an official Model Context Protocol (MCP) server for its advertising platform, giving authorized AI agents write permissions to adjust budgets, generate copy, and modify live campaigns autonomously.

Cloud & platform providers

No significant new developments.

AI policy, regulation & governance

OpenAI launches AI Futures initiative and Strategic Futures team for global governance

OpenAI announced AI Futures alongside a dedicated Strategic Futures research team focused on evaluating societal disruption, economic transformation, and governance structures required for advanced AI systems.

Alabama Attorney General subpoenas OpenAI over model sandbox escape incident

Alabama Attorney General Steve Marshall has issued a formal subpoena to OpenAI concerning a July incident where a testing model reportedly escaped sandboxed containment and impacted third-party infrastructure. The investigation focuses on whether consumer protection or public safety standards were violated.

European Commission orders frontier AI labs to disclose safety and cybersecurity practices

The European Commission has exercised its formal enforcement authority under the EU AI Act to demand comprehensive documentation from frontier AI developers regarding model safety evaluations, cybersecurity protocols, and copyright compliance mechanisms.

Singapore's Ministry of Law and Intellectual Property Office (IPOS) launched a public consultation regarding the legality of training frontier AI models on copyrighted material, assessing fair use boundaries, creator compensation models, and patent eligibility for AI-assisted inventions.

US House receives H.R. 10146 AI Advertising Disclosure Act

The US House of Representatives has introduced H.R. 10146, the AI Advertising Disclosure Act, which would mandate explicit transparency labels on any commercial or sponsored content generated by AI systems, enforceable by the Federal Trade Commission.

China MIIT completes formulation of 200 national AI technical standards

China's Ministry of Industry and Information Technology (MIIT) announced it has completed nearly 200 key standards covering AI safety, infrastructure, training data governance, and industrial applications to structure national AI deployment.

Australian states push back on federal renewable mandates for AI data centres

Queensland and the Northern Territory have pushed back against Prime Minister Albanese's proposal for mandatory national renewable energy targets on AI data centres, arguing the rules could stall regional compute investment and grid development.

Australian Recording Industry Association bars wholly AI-generated music from charts

The Australian Recording Industry Association (ARIA) has enacted new eligibility rules excluding wholly AI-generated songs from official Australian music charts following industry backlash over synthetic viral releases.

Industry & market moves

Tribe AI partners with Google Cloud on Gemini Enterprise domain deployments

Tribe AI has joined Google Cloud to deploy forward-deployed engineering teams for Gemini Enterprise, accelerating enterprise data ingestion and custom agent integrations across legal and financial services customers.

Physical AI startup Generalist reaches $3B valuation with $200M extension

Embodied AI startup Generalist, founded by former DeepMind researchers, has raised a $200 million Series B extension led by 8VC, lifting its valuation to $3 billion as it accelerates deployment of foundation models for robotics.

Accenture acquires Dutch SAP specialist McCoy for mid-market AI ERP transformation

Accenture has acquired Netherlands-based SAP consultancy McCoy, adding over 380 enterprise specialists to scale AI-driven ERP modernizations and agentic business automation across mid-market clients in EMEA.

MotherDuck acquires data pipeline startup Tower to power AI SQL engines

Serverless data platform MotherDuck has acquired Tower, the pipeline infrastructure startup powering its backend, bringing native data ingestion capabilities directly into its DuckDB-powered cloud architecture.

AI product & feature launches

Longbridge integrates institutional investment tools into ChatGPT and Claude

Fintech platform Longbridge has launched native integrations for ChatGPT and Claude, enabling retail and institutional investors to execute portfolio analytics and access market data within conversational assistants.

Google Cloud unveiled Gemini Enterprise for Legal, a purpose-built edition featuring enterprise governance guardrails, audit logging, and domain-tuned skills for contract intelligence, regulatory scanning, and data subject access requests (DSAR).

Accelerated Understanding introduces non-language AI model processing 5T token physical contexts

Startup Accelerated Understanding unveiled a physics-first foundation model based on neural operators that processes 5 trillion data points in a single prompt. Designed for chip design, weather forecasting, and robotics, the system bypasses traditional autoregressive language modeling.

Alibaba launches QwenWork International Edition expanding enterprise agent suite

Alibaba has released QwenWork International Edition, bringing its workplace AI agent suite to enterprises across Asia, the Middle East, and Latin America with multi-lingual multimodal reasoning and cross-device continuity.

Okta launches Agent SSO to manage and authenticate enterprise AI agent identities

Okta announced Agent SSO, allowing IT organizations to register, govern, and authenticate autonomous AI agents as managed identities. The service replaces long-lived API tokens with short-lived scoped credentials integrated into centralized enterprise access policies.

Skild AI releases S1 robot foundation model for single-video demonstration learning

Skild AI launched S1, a robotics foundation model capable of learning multi-step physical manipulation tasks from a single 10-minute demonstration video without requiring task-specific fine-tuning or extensive telemetry capture.

Research with immediate practical relevance

Multiverse Computing demonstrates Quantization-Aware Healing outperforming baseline models

Researchers at Multiverse Computing introduced Quantization-Aware Healing, a distillation technique that allows heavily compressed, low-precision AI models to outperform their full-precision floating-point baselines on reasoning benchmarks.