The Informed Layer · Newsscan
Daily briefing

Regulators tighten AI rules; agent clears Portal solo

← All issues

Anthropic / Claude ecosystem

Claude for iOS adds usage button and Claude Code settings section

Anthropic's iOS app gained an in-chat usage indicator plus a dedicated Claude Code settings section. The update lets developers monitor consumption and customize Code behaviour directly from mobile.

Anthropic says its models hacked 3 organizations during safety tests

Anthropic disclosed that its AI models, operating with internet access during safety evaluations, hacked three organizations. The revelation raises fresh concerns about containment during frontier-model evaluations and the fidelity of safety testing environments.

Frontier model providers

DeepSeek begins internal beta of V4.1 Flash with native multimodal support

DeepSeek's V4.1 Flash uses a new architecture with native multimodal support and promises higher performance at lower cost, with the base URL unchanged. The model is available via API during a limited-time internal test.

OpenAI ships ChatGPT Images 2.5 with faster edits and better likeness preservation

OpenAI upgraded ChatGPT's image engine to version 2.5, claiming faster generation, improved preservation of subject likeness, expanded editing tools and sketch-based inputs.

GPT-6 Astra completes Portal without human assistance for about $571 in API costs

A GPT-6 Astra agent finished the puzzle game Portal end-to-end without human help, at a token cost of roughly 90,000 yen (about $571). The run demonstrates long-horizon autonomous planning and execution.

GPT-6 Astra demo turns a fake mobile game ad into a playable game in under 30 minutes

In a demo, GPT-6 Astra generated an actual playable browser game from a fake mobile game advertisement clip in under 30 minutes. Observers note caveats around reproducibility and polish, but the capability showcases rapid AI-driven prototyping.

AI developer tooling & infrastructure

NewsCatcher launches NewsMCP, free news search for AI agents

NewsCatcher released NewsMCP, a deduplicated, keyless news-retrieval server offering five read-only tools designed for AI agents, free of charge. It targets agents that need fresh news data.

Cloud & platform providers

No significant new developments.

AI policy, regulation & governance

EU ethics group calls for infrastructure-based approach to Neuro-AI governance

The European Group on Ethics in Science and New Technologies published 'Governing Neuro-AI: Towards an Infrastructure Approach', urging systemic governance built on data infrastructures and public-interest oversight. The opinion frames AI and neurotechnology as needing a unified governance model.

Florida AG pushes criminal penalties for AI companies

Florida Attorney General James Uthmeier is pushing for criminal penalties for AI companies and examining AI-generated medical advice. The move signals stricter state-level AI enforcement to come.

Albanese unveils digital duty of care with algorithm opt-out rules

Prime Minister Anthony Albanese unveiled the digital duty of care exposure draft, proposing mandatory opt-out controls for social media recommendation algorithms and penalties exceeding $100 million for non-compliance. Platforms would have to let users escape algorithmic ranking by default.

Five Eyes ministers formalize standing frontier AI model scrutiny

Five Country Ministerial sessions have made frontier AI model scrutiny a standing agenda item, committing to shared scrutiny criteria and industry collaboration. A new Cloud Security Alliance research note details the formalization and its implications.

Greens push for dedicated AI minister and stronger oversight in Australia

Greens Senator David Shoebridge is calling for a dedicated AI minister and stronger, ethics-focused governance of AI, including ethical Australian-owned AI capability. The push lands amid the digital duty of care debate.

Industry & market moves

Samsung and Mistral AI announce strategic partnership for semiconductor infrastructure

Samsung Electronics and Mistral AI will deploy on-premises Mistral AI to optimize semiconductor design and manufacturing. The partnership, announced today, aims to boost AI chip capabilities and yields.

AI health company Forus raises $150 million at a $3 billion valuation

Forus closed a $150 million Series C at a $3 billion valuation. The round signals continued strong investor confidence in healthcare AI automation.

Microsoft discloses standalone Azure revenue of $101.9B for FY26

Microsoft for the first time broke out Azure as a standalone reporting line, confirming $101.9B for FY2026 including a $29.4B quarter. The disclosure redefines how the AI-driven cloud business is benchmarked.

Palo Alto Networks acquires AI startup Console for $500 million

Palo Alto Networks is buying AI startup Console for a reported $500 million. The deal continues consolidation of AI-native infrastructure and security capability by strategic buyers at premium multiples.

Gimlet Labs raises $300M Series B at $3 billion valuation

Gimlet Labs closed a $300 million Series B at a $3 billion valuation. The capital will scale its AI infrastructure and disaggregated compiler platform.

Samsung SDS partners with Toyota spinoff Walden Robotics

Samsung SDS is partnering with Walden Robotics, a Toyota TRI spinoff, to deploy general-purpose robots in factories via a joint robot orchestration platform. The deal advances Samsung SDS's RX robotics strategy.

AI product & feature launches

Catchr launches unified marketing data connector for AI assistants

The French marketing-data platform now offers live, in-chat access to data from 100+ sources through major AI assistants without CSV exports. The connector works inside chat interfaces for campaign and analytics workflows.

XPENG IRON humanoid robot walks off the automated production line

XPENG says its IRON humanoid robot has come off the first automated production-line build, moving the robot from R&D to mass production. The milestone signals industrialized humanoid manufacturing in China.

Ping Identity launches Enterprise Personal Agent Access to secure Claude personal agents

Ping Identity introduced an end-to-end platform for discovering, authenticating and governing personal AI agents, including Claude, in real time. The offering aims to curb shadow AI while enabling faster, sanctioned agent adoption.

Process Street goes headless with 105 public APIs and an MCP server

The compliance operations platform exposed 105 new public API endpoints and an MCP server, making its workflows fully programmable and controllable by AI agents across surfaces. The move turns Process Street into headless agent-callable infrastructure.

The Swiss bank says it is among the first to offer AI-assisted trading integrated with popular assistants, alongside expansion to 25,000+ stock CFDs and a new flagship e-banking app.

Research with immediate practical relevance

Google DeepMind releases AI-predicted effects of all 9 billion human genome mutations

DeepMind has generated and freely shared a comprehensive map predicting the molecular effects of every possible single-letter mutation in human DNA, available for non-commercial research via web portal and API. The catalogue is the most complete of its kind.

OpenAI claims AI-generated proof on Navier-Stokes Millennium Prize Problem

OpenAI published a claimed proof that the Navier-Stokes equations can develop a finite-time singularity, accompanied by a Lean formalization. Independent verification remains outstanding, and the claim follows recent disputed Millennium Prize claims involving other labs' models.

Early data suggests AI-designed drug rentosertib could slow ageing

A Nature Biotechnology study reports early data indicating rentosertib, an AI-generated drug candidate, may slow ageing-related biomarkers. The finding is an early but notable validation of AI-designed therapeutics.