The Informed Layer · Newsscan
Daily briefing

Anthropic flags model risk; AI agents act autonomously

← All issues

Anthropic / Claude ecosystem

Anthropic details unreleased Model 2 and new alignment concerns in latest AI risk report

Anthropic disclosed an internal, more capable model called Model 2 in its latest risk report, along with updated alignment risk estimates and expanded internal use. The model is not planned for external release.

Samsung is reportedly using Claude to speed up chip design

Samsung Electronics' System LSI division is reported to be using Anthropic's Claude Code to accelerate chip verification tasks substantially while maintaining human oversight.

Frontier model providers

Grok 4.6 created a 3D shooter autonomously in 48 hours

xAI's Grok 4.6 autonomously built a complete 3D shooter in 48 hours, demonstrating rapid autonomous software generation.

You can now turn off Google Gemini's visible watermarks

Google added a user-controllable toggle to hide visible AI watermarks in Gemini while retaining invisible markers for verification.

WhatsApp will use AI to protect users from online scams

WhatsApp is testing an on-device AI scam-detection feature that alerts users to suspected scams within chats while preserving end-to-end encryption.

Cohere North Micro Vision Instruct: Apache 2.0 release

Cohere released North Micro Vision Instruct under the Apache 2.0 license, providing an open vision-language model for broad commercial use.

DeepSeek V4 Pro: better benchmarks, higher prices, and a bigger AI ambition

DeepSeek introduced V4 Pro with stronger engineering benchmarks while some V4 family API prices rose more than 10x as demand strained capacity.

AI developer tooling & infrastructure

LLM 0.20 release adds reasoning chain tracking, OpenAI response objects, and server-side tools

LLM 0.20 introduces inference chain tracking, access to raw OpenAI response objects, and server-side tooling for enterprise LLM APIs.

Cloud & platform providers

No significant new developments.

AI policy, regulation & governance

Governor Abbott announces Stack Infrastructure, Anthropic, and Nightpeak Energy commit to Texas data center standards

Texas Governor Abbott reported that Stack Infrastructure, Anthropic, and Nightpeak Energy have committed to comply with statewide data center guardrails covering electric grid, water use, and local impacts.

Korea unveils AI ethics principles

Korea formally announced AI Ethics Principles with three core values and seven principles to guide desirable AI development, following earlier consultation.

Insurance regulators get schooled on AI governance

Insurance regulators are developing a common evidentiary framework to assess insurer use of AI, covering data lineage, validation, monitoring, and third-party vendor oversight.

OpenAI alerted FBI after man told ChatGPT he planned to kill his ex

OpenAI flagged explicit violent threats made in ChatGPT to law enforcement, leading to a criminal case and probation.

White House AI safety reviews could expand to open models

The administration may broaden voluntary safety reviews to frontier open models, signaling a capability-based regulatory approach rather than one based only on model openness.

Industry & market moves

HP Sarvam AI deal: Indian language voice tech could revolutionise PC use

HP plans to integrate Sarvam's voice and language AI capabilities into devices, beginning with Kivi, enabling multilingual AI-powered PC experiences.

Aizen enters collaboration with San Diego biopharma to advance oral biologics

Aizen Therapeutics will use its DaX AI-driven peptide design platform to create oral biologics, with target milestones potentially worth up to $100 million per target.

An AI running an SF store fired an employee for the first time

Luna, an AI manager at Andon Market in San Francisco, fired a human employee, marking the first known instance of AI-driven employment termination.

Burson acquires cognitive AI company Limbik

Burson acquired Limbik to embed cognitive-AI into its predictive intelligence platform Decipher and accelerate product development across its suite.

Buildertrend furthers its AI ambitions with acquisition of BizJet AI

Buildertrend acquired early-stage AI startup BizJet AI and brought its co-founders onto the team to expand its AI strategy.

EMERGING and Promethean launch $300M Experience Fund to back AI for hospitality

A new $300 million AI-for-hospitality fund, The Experience Fund, will back AI, IP, and automation in hospitality and experiential entertainment.

AI product & feature launches

KISTI unveils 'AI Scientist' that generates research hypotheses

Korea's KISTI introduced an AI Scientist system to generate research hypotheses from scientific literature, alongside Web Agent and KONI-Forge, to accelerate discovery.

China's Z.ai launches model it says rivals Anthropic's Mythos

Z.ai launched GLM-5.3, claiming advanced coding and security capabilities that rival major competitors including Anthropic's Mythos.

Alibaba Cloud launches Zhenwu M890 supernode for commercial use

Alibaba Cloud has started commercial service for its Zhenwu M890 supernode, targeting larger and faster AI infrastructure deployments.

Fiserv and Stuut partner on agentic AI for enterprise AR

Fiserv and Stuut are integrating agentic AI into Fiserv's payments rails to automate B2B accounts receivable, targeting more than $2 billion in invoice automation.

Research with immediate practical relevance

Two humanoid robots just learned to turn a skipping rope together

A framework called Marope uses multi-agent reinforcement learning to coordinate two Unitree G1 humanoids in cooperative rope-skipping, validated in simulation and real-world tests.

Alignment with experimental data improves protein generative modeling

Researchers introduced ProteinDPO, a method that aligns protein generative models with experimental fitness data to improve stability while preserving function.