The Informed Layer · Newsscan
Daily briefing

Policy and security tighten; models and robots advance

← All issues

Anthropic / Claude ecosystem

Claude safety mechanism failure reportedly deletes developer's 700GB home directory

A safety downgrade in Claude led to a variable reuse bug during cleanup, causing catastrophic deletion of a developer's 700GB home directory. The incident highlights a rare failure mode in AI-assisted coding safety mechanisms.

Frontier model providers

Meta prices AI agent that orders DoorDash at $199.99 per month

Meta is reportedly charging $199.99 per month for an AI agent that can order from DoorDash and perform other consumer tasks. The pricing signals Meta's push into monetised autonomous consumer agents.

AI developer tooling & infrastructure

New phi-guard-mcp tool prevents PHI leaks in AI-generated code

phi-guard-mcp is an open-source local scanning tool that prevents protected health information from leaking through AI-generated code. It addresses a specific data-loss risk in healthcare AI workflows.

Huawei Cloud launches CodeArts Agent in Singapore

Huawei Cloud introduced CodeArts Agent, an AI coding assistant with full lifecycle support for developers and enterprises, in Singapore. The launch targets the APAC market.

Cloud & platform providers

Cloudflare launches Adaptive Intelligence to rewrite bot rules on the fly

Cloudflare introduced Adaptive Intelligence, a live detection engine that continuously rewrites short-lived bot rules to adapt to evolving attack techniques. It automates security response without manual rule updates.

AI policy, regulation & governance

California Legislature passes independent AI safety verification bill

The California Legislature passed a bill requiring independent verification of AI safety measures. The bill now moves toward the governor and would create new compliance obligations for AI developers and deployers.

TH-AI Passport goes live for 5 million Thais

Thailand launched TH-AI Passport, a government-backed program providing access to free premium AI tools for 5 million users. The launch was announced on Thaiger's platform.

FSB warns AI-driven cyber risk is top global financial stability concern

The Financial Stability Board, with Andrew Bailey, warned that AI-enabled cyber risk could rapidly alter financial stability and requires global resilience and safe model release practices. The warning highlights systemic risk from AI attacks.

Australia's eSafety regulator says smart glasses should automatically blur faces

Australia's eSafety Commissioner is pushing for smartglasses to automatically blur faces while filming and provide clear recording indicators. The safer design guidelines extend earlier scrutiny of wearable cameras.

Industry & market moves

DGIST and Yuhan partner to accelerate AI drug development

South Korea's DGIST and Yuhan are partnering to establish an AX Joint Research Lab targeting AI-driven drug development across research, clinical validation, and commercialisation. The collaboration aims to shorten drug discovery cycles.

HITAI acquires YC-backed Quell to boost AI sports motion tracking

London-based HITAI acquired fitness gaming studio Quell to strengthen its sports motion-tracking platform with camera-less motion recognition and real-time performance models. The deal adds technology and IP.

Asus launches physical AI business group targeting humanoid robots

Asus rebranded its AIoT arm as the Physical AI Solutions Business Group and identified humanoid robots as a core growth area, signalling a shift from EMS manufacturing to robotics.

SenseTime spinout Sunrise reportedly raises RMB2 billion at RMB20 billion valuation

Sunrise, a SenseTime spinout, has reportedly raised RMB2 billion at a RMB20 billion valuation. The funding will support its AI business expansion.

Agentrys raises $24.5 million for AI chip design automation

Agentrys secured $24.5 million in funding to accelerate agent-based AI design automation for semiconductor design and verification. The investment supports autonomous EDA workflows.

CrowdStrike CEO says Anthropic Mythos helped drive record quarterly revenues

CrowdStrike CEO George Kurtz acknowledged that Anthropic's Mythos model contributed to record quarterly revenues while warning that AI agents are accelerating cyber threats. The statement links frontier AI models to security market growth.

AI product & feature launches

Claude-Red v0.3.0 released for security applications

Claude-Red v0.3.0 is now available, focused on security and red-teaming capabilities. The release adds new features for adversarial testing and security workflows.

iFlytek to release Spark X2.5 general model on September 7

iFlytek will launch its Spark X2.5 293B foundation model on September 7, with upgraded coding, agent, and core capabilities for automotive and smart hardware. The release targets edge and cloud scenarios.

Runway introduces Solaris, an interface world model that renders UIs frame-by-frame

Runway launched Solaris, the first model in its Interface World Models family, which renders and updates user interfaces frame-by-frame in real time. The tool targets creators building interactive AI experiences.

Uncensored AI Studio packs local image generation, LLM, and voice on a USB drive

A new portable USB-based AI suite bundles image generation, LLM, and voice tools with automatic hardware acceleration detection. It targets users seeking local AI without cloud dependencies.

Nothing OS 5.0 goes official with Android 17 and expanded AI features

Nothing OS 5.0 launches with Android 17, over 80 changes, and a broader AI push including on-device intelligence improvements and MCP integration. The update rolls out as a staged beta.

VerSe Innovation unveils SparkStation AI content OS to cut brand film production costs by up to 90%

VerSe Innovation launched SparkStation, an AI-powered Content Operating System that consolidates content production workflows. It claims to cut 60-second brand film production costs by up to 90% and shorten timelines to about 24 hours.

Research with immediate practical relevance

Google launches Planetary Prediction Engine to automate global models via Earth AI

Google Research announced a planetary prediction engine that executes the full geospatial prediction workflow from natural-language queries. It reduces model-building time from weeks to minutes.