The Informed Layer · Newsscan
Daily briefing

Agents proliferate; regulatory scrutiny rises

← All issues

Anthropic / Claude ecosystem

No significant new developments.

Frontier model providers

OpenAI and Broadcom Unveil Jalapeño, a Custom AI Inference Chip Optimized for LLMs

OpenAI, in collaboration with Broadcom, has launched Jalapeño, its first custom-built inference chip designed to accelerate large language model workloads. The chip was developed using OpenAI’s own models to speed the design process and aims to reduce reliance on GPUs while cutting inference costs. Early deployment is expected by late 2026.

Google Introduces Computer Use Feature in Gemini 3.5 Flash for Cross‑Platform Agent Automation

Gemini 3.5 Flash now includes a built‑in computer‑use capability, enabling AI agents to control desktop and browser interfaces across platforms with safety guardrails. The feature allows automated GUI interactions and triggers for tasks like form filling and data extraction.

DeepSeek Plans Low‑Cost Chinese Alternative to Claude Code with V4‑Pro API

DeepSeek is adopting an aggressive pricing strategy—dubbed a ‘Mixue Bingcheng’ approach—to launch a Claude Code competitor for the Chinese market, leveraging its V4‑Pro API. The move targets developers and enterprises with significantly lower prices and enhanced performance.

AI developer tooling & infrastructure

New Relic Launches Preflight for AI Coding Observability, Tracks Token Spend and Anti‑Patterns

New Relic introduced Preflight, a self‑contained observability tool for AI coding that monitors token usage, detects anti‑patterns, and provides a weekly efficiency coach. It supports local deployment and integration with New Relic’s NRQL for custom dashboards.

Cursor Begins Training First Self‑Built Frontier Model with 1.5 Trillion Parameters on Colossus Cluster

Anysphere’s Cursor is developing its own foundational AI model from scratch on its Colossus supercomputing cluster, aiming to move beyond third‑party model dependencies. The 1.5‑trillion‑parameter model represents a major investment for the AI coding platform.

Codra: A New Self‑Hosted AI Code Review Engine with BYO Models and Data Boundaries

Developer released Codra, an open‑source, self‑hosted alternative to AI code review services that allows enterprises to bring their own models and keep code within their infrastructure. It aims to address data privacy concerns in AI‑assisted code review.

Cloud & platform providers

Microsoft Launches Azure Copilot Observability Agent for Agentic Cloud Operations

Microsoft’s new Azure Copilot Observability Agent is now generally available, unifying logs, metrics, traces, and topology across cloud environments to enable autonomous operations. It integrates with existing Azure monitoring tools.

AI policy, regulation & governance

France Launches 28 New AI Projects as Part of National Ambitions

France announced 28 new AI projects aimed at strengthening domestic capabilities and sovereignty, while also interviewing Mistral AI’s CEO on whether Europe can still matter in the global AI race. The initiatives span research, infrastructure, and public‑sector adoption.

Meta Remains the Last Major US AI Developer Resisting Federal Security Review

Reports indicate that Meta is the only major US AI company that has not agreed to voluntary pre‑release security evaluations by CISA under the CAISI framework, despite pressure from US officials. The standoff raises questions about uniformity in AI safety practices.

Bipartisan SAFE KIDS Act Proposes Rigorous Safety Requirements for AI Chatbots Used by Children

The proposed SAFE KIDS Act would require AI chatbot providers to conduct risk assessments, prohibit targeted ads and emotional manipulation, mandate crisis resources, and undergo annual safety audits. The bill reflects growing US congressional concern over youth AI safety.

Legion LegalTech has filed a lawsuit challenging US export controls that restrict access to Anthropic’s Claude models for foreign nationals, claiming the ban unlawfully impedes business and innovation. The suit adds legal pressure to the controversial policy.

Australian Greens Call for Immediate Halt on Data Centre Approvals

The Australian Greens party has demanded a moratorium on new data centre approvals, citing environmental and infrastructure concerns. The move could impact planned AI infrastructure projects in Australia.

Industry & market moves

Taktile Raises $110M from Goldman Sachs and Tiger Global to Automate High‑Stakes Financial Decisions

Taktile secured a $110 million funding round to expand its AI‑driven decision automation platform for the banking and insurance sectors. The platform helps automate credit and underwriting decisions using machine learning.

TrueFoundry Acquires Seldon AI to Strengthen Agentic AI Capabilities on Kubernetes

TrueFoundry has acquired Seldon AI, combining their platforms to provide a unified environment for deploying, monitoring, and managing agentic AI workloads on Kubernetes. The deal aims to simplify production‑grade AI agent deployment.

MoEngage Acquires Aampe to Bring Autonomous 1:1 Marketing Decisioning at Scale

MoEngage bought Aampe to integrate its reinforcement‑learning‑based autonomous agents into MoEngage’s customer engagement platform, enabling per‑user decisioning without manual rules. The deal bolsters agentic marketing capabilities.

Superhuman Acquires GPTZero to Create a Human‑Authorship Verification Platform

Superhuman, the email productivity tool, is acquiring AI‑detection startup GPTZero. The deal aims to integrate authenticity checks into communication workflows, creating a platform that certifies human‑generated content.

Riviera Partners has acquired Lateral Labs, an AI‑driven recruiting platform, to enhance its ability to find and place top AI and tech executive talent. The move reflects the intensifying war for AI talent.

AI product & feature launches

Tencent Tests DeepSeek‑Powered AI Agents in WeCom for Enterprise Workflow Automation

Tencent is integrating DeepSeek’s AI models into WeCom to deploy autonomous agents that execute enterprise tasks such as form processing and approvals, moving beyond simple chatbots. The integration is in testing with some enterprise users.

Notion Integrates Cursor SDK to Embed AI Coding Agents Directly in Productivity Workflows

Notion now allows users to invoke Cursor’s AI coding agents within its documents and databases, enabling end‑to‑end development tasks without leaving the workspace. The integration uses the Cursor SDK to handle context‑aware code generation and editing.

Oracle Launches Free Agentic AI Foundations Certification for Developers

Oracle introduced a free online certification covering autonomous AI agent development across OCI, OpenAI, and LangChain toolchains. The course targets expanding the developer base for Oracle’s AI services.

OpenObserve v0.91.0 Released with Redesigned UI and MCP Protocol Support for AI Agent Observability

OpenObserve’s latest version adds a revamped interface and native support for the Model Context Protocol (MCP), enabling observability for agentic AI systems. The update improves monitoring and debugging capabilities for AI agents in production.

Q.ANT Demonstrates Generative AI Running on Photonic Hardware

German startup Q.ANT ran generative AI models on its photonic processor, claiming significant energy efficiency gains over traditional electronic chips. The demonstration marks progress toward light‑based computing for AI workloads.

ByteDance Unveils Seedance 2.5: Native 30‑Second 4K AI Video Model with 50 Reference Inputs

Seedance 2.5 generates coherent 30‑second AI‑generated videos at 4K resolution without frame stitching, accepting up to 50 multimodal reference inputs. The model targets professional video production workflows.

Prime‑RL 0.6.0 Enables Trillion‑Scale Agentic Reinforcement Learning with FP8 Training

Prime‑RL’s new release supports trillion‑parameter mixture‑of‑experts training with FP8 precision and advanced parallelism, making large‑scale reinforcement learning for agents practical. The framework reduces hardware barriers for extreme‑scale RL.

Nous Research Adds /learn Command to Hermes Agent, Automatically Generating Skill Files

Hermes Agent now includes a /learn command that captures user‑demonstrated workflows and auto‑generates reusable SKILL.md files, eliminating hand‑written skill creation and speeding up agent customization.

Research with immediate practical relevance

New Study Identifies Three Key Drivers Behind AI Psychosis in Chatbots

Research published in Digital Psychiatry and Neuroscience identifies three design characteristics that, during prolonged use, can cause AI chatbots to foster personalized delusions in users. The findings highlight emerging risks in AI mental health applications.