The Informed Layer · Newsscan
Daily briefing

Agents surge; governance frameworks tighten

← All issues

Anthropic / Claude ecosystem

Anthropic Adds 28 Security and Compliance Integrations to Claude Enterprise

Anthropic introduced 28 new security and compliance platform integrations to Claude Enterprise, including a Cloud Compliance API. This allows customers to govern and monitor AI usage across their existing security stacks.

Claude Code Gets Free Security Plugin to Detect and Fix Vulnerabilities

Anthropic released a security‑guidance plugin for Claude Code that automates vulnerability detection and remediation during development. It identifies issues as code is written and at commit time.

Frontier model providers

OpenAI Details Self‑Improving Tax Agents with Codex

OpenAI published a guide on building self‑improving tax AI agents using Codex. The system uses bounded task environments and production traces to automatically fix errors in tax workflows.

Google SynthID Becomes Default Watermark for AI Media

Google’s SynthID is emerging as the default watermarking standard for AI‑generated images, video, audio, and text, with major AI players integrating it for provenance and compliance.

Meta Plans First Paid ‘Meta AI Subscription’ Tiers Under Meta One

Meta is reportedly planning paid AI subscription tiers—Meta One Plus ($7.99/mo) and Meta One Premium ($19.99/mo)—alongside app‑specific plans, aiming to monetize Meta AI without altering the ad‑based model.

Google Moves Users from Open‑Source Gemini CLI to Closed‑Source Antigravity CLI

Google is retiring its open‑source Gemini CLI for most users in favor of the closed‑source Antigravity CLI, imposing usage limits on non‑enterprise tiers. Enterprise users retain access.

Meta Unveils Ad AI Connectors for Conversational Campaign Management

Meta introduced AI‑driven conversational tools for ad creation and management, enabling marketers to set up campaigns via AI assistants and potentially automate execution.

OpenAI Targets Smaller Advertisers to Monetize ChatGPT Ads

OpenAI is expanding its ChatGPT ad offerings to smaller advertisers, aiming to build a more measurable ad platform beyond large brands. The push follows the earlier Ads Manager beta.

AI developer tooling & infrastructure

42Crunch AI Coding Plugins Unlock Agentic DevSecOps with Claude Code

42Crunch integrated its API security testing with Claude Code, enabling real‑time detection and automatic remediation of API vulnerabilities within AI‑driven development. The plugin creates a continuous detect‑and‑fix loop.

Giggso Launches Raven, Andie, and AIRTaaS for Enterprise AI Governance

Giggso introduced three AI governance offerings—Raven, Andie, and AIRTaaS—designed to reduce comprehension debt and improve discipline, reasoning, and security in enterprise AI adoption.

DeepSWE Benchmark Released to Evaluate AI Coding Agents on Long‑Horizon Tasks

Serena Ge released DeepSWE, a benchmark for AI coding agents that tests complex, long‑horizon engineering tasks such as repository‑level feature development. It aims to set a new standard for agentic coding evaluation.

Sensedia Advances AI Gateway for Enterprise Control in Agentic Era

Sensedia announced general availability of its AI Gateway, designed to govern autonomous agents and API interactions with token observability and policy enforcement.

Exabase M‑1 Memory Engine Achieves Top Score on LongMemEval with Smaller Model

Exabase’s M‑1 memory engine scored highest on the LongMemEval benchmark using a cheaper, smaller model, demonstrating that production‑ready memory architecture can boost reasoning with less compute.

EAGLE 3.1 Fixes LLM Inference Drift, Doubles Throughput

EAGLE 3.1, a speculative decoding technique for vLLM, resolves attention drift and delivers up to 2× throughput improvement in production LLM inference.

Bugcrowd Releases RL Environments for AI Security Training

Bugcrowd launched reinforcement learning environments that train AI models on locating, exploiting, patching, and auditing software vulnerabilities, using open‑source code without customer data.

Cloud & platform providers

Amazon Sells AI Shopping Assistant Tech to Retailers via AWS

Amazon deployed its first production‑ready retail AI assistant, built with Amazon Bedrock AgentCore, for Kate Spade. The service is available to other retailers through AWS.

Amazon SageMaker Unified Studio Adds Feature Store Interface

Amazon SageMaker Unified Studio now includes an interactive, code‑free interface for managing feature groups in Feature Store, enabling collaborative ML feature management.

Hugging Face Releases 3D‑Printable LeRobot Humanoid

Hugging Face launched LeRobot Humanoid, an open‑hardware, 3D‑printable robot platform with a full software stack for rapid simulation‑to‑physical testing and learning experiments.

AI policy, regulation & governance

Vatican Partners With Anthropic to Shape AI Ethics

The Vatican and Anthropic announced a partnership to develop an external moral framework for AI governance. The collaboration aims to address ethical concerns amid rapid AI advancement.

Anthropic and Vatican Launch First AI Encyclical

Anthropic and the Vatican co‑released the first AI‑focused encyclical, Magnifica Humanitas, providing a high‑level moral and ethical framework for AI development.

Meta Lawsuit Over AI Scam Ads Targets Core Ad Model

A new lawsuit alleges Meta’s AI‑driven ad system facilitates scam ads that prey on vulnerable users, focusing on the platform’s ad targeting algorithms rather than privacy or competition issues.

NSA Warns MCP Poses Cyber/Operational Risks for Banks

The NSA issued a warning that the Model Context Protocol (MCP), used for AI automation, could introduce significant cyber and operational risks if not properly safeguarded, especially for banks deploying autonomous AI.

Illinois AI Safety Bill Advances with Third‑Party Audits and Incident Reporting

The Illinois Senate passed Senate Bill 315, the AI Safety Measures Act, requiring AI system audits and incident reporting. The bill aligns with growing state‑level AI regulation.

EU Commission Opens Public Discussion on High‑Risk AI Classification Draft

The European Commission launched a public consultation on draft guidelines that clarify which AI systems qualify as high‑risk under the EU AI Act. Comments are open until June 23, 2026.

Trump to Sign Executive Order Creating Pre‑Release Vetting for AI Models

President Trump is expected to sign an executive order establishing a formal vetting system for frontier AI models before public release, focusing on security evaluation.

South Korea Becomes Third Country to Join OpenAI GTAC for AI Security

South Korea’s Ministry of Science and ICT joined OpenAI’s Global Threat Analysis Center (GTAC), partnering on a cyber action plan and institutional AI security measures.

South Africa Delays AI Policy After Fake Citations Crisis

South Africa postponed its national AI policy to 2027 after fabricated references in policy documents undermined government credibility. The delay casts uncertainty on the country’s regulatory timeline.

Michigan Bill Proposes AI Pilot Program for State Government

House Bill 5899 in Michigan would create an AI oversight board and a pilot program to study AI use across state agencies, with defined setup and ongoing costs.

FDA Seeks Public Input on AI in Early‑Phase Clinical Trials

The FDA opened a comment period on a pilot program, AI‑Enabled Optimization of Early‑Phase Clinical Trials, seeking input by May 29 on using AI to streamline trial design.

APRA Issues Binding AI Risk Expectations for Australian Banks

Australia’s prudential regulator APRA formalized binding AI risk expectations across four domains, with escalation pathways for non‑compliance. The letter tightens governance of AI in financial institutions.

Industry & market moves

OpenAI Partners with Brazilian News Groups for Real‑Time Responses

OpenAI partnered with Grupo Folha and Grupo UOL to integrate their news content into ChatGPT, enabling real‑time, attribution‑backed Brazilian news in AI responses.

Cohere and Mila Partner to Advance Quebec French Language AI

Cohere is collaborating with Mila, the Quebec AI research institute, to adapt AI evaluation and training for Quebec’s French‑language cultural context. The partnership focuses on high‑context multilingual environments.

BNP Paribas and Mistral Team to Defend Against Mythos‑Like AI Threats

BNP Paribas is partnering with Mistral AI to bolster cybersecurity against emerging AI‑driven threats, including those potentially enabled by advanced models like Mythos. The bank aims to benchmark AI‑powered defenses.

DeepSeek Builds Code Harness Team for Desktop Agent Product

DeepSeek is forming a new Harness team to develop a desktop coding agent, signaling a strategic push into the AI coding assistant market. The initiative follows a $7 billion funding round.

Greenhouse Acquires Ezra AI Labs for Conversational AI in Hiring

Greenhouse acquired Ezra AI Labs, bringing structured, voice‑led interviewing to its hiring platform. The deal adds conversational AI to the initial candidate screening stage.

Ubase to Acquire Voice AI Startup ReturnZero for Counselling Agent

South Korean BPO firm Ubase plans to acquire voice AI startup ReturnZero, with the goal of launching an end‑to‑end AI counselling agent in July. The move internalises voice AI capabilities.

SoftBank’s Homegrown AI Project Garners Top Japanese Manufacturers

SoftBank is courting around 30 Japanese manufacturers to invest in a domestic AI venture, aiming for a major national AI collaboration. The project could rival foreign‑dominated AI infrastructure.

Terra Quantum to Go Public via $3.5B SPAC Merger

Terra Quantum AG and Axiom Intelligence Acquisition Corp 1 announced a definitive business combination agreement at a $3.5 billion equity valuation, with the combined entity to list on Nasdaq.

Liminatus Pharma to Acquire InnocsAI in $320M CAR‑T Oncology Deal

Liminatus Pharma announced a $320 million share‑based deal to acquire InnocsAI, expanding its CAR‑T oncology pipeline with AI‑enabled capabilities.

OpenRouter Doubles Valuation to $1.3B in Series B

AI model gateway OpenRouter raised $113M in a Series B round, more than doubling its valuation to $1.3 billion in one year, reflecting strong demand for multi‑model API platforms.

IntBot and Certis Partner on Enterprise Physical AI in Singapore

IntBot and Certis Group announced a partnership to deploy socially intelligent, enterprise‑grade robots in Singapore, targeting public‑facing environments.

Qualcomm Lands ByteDance AI Chip Deal for Data Centers

Qualcomm secured a deal to supply application‑specific AI chips to ByteDance’s data centers, marking its expansion from smartphone processors into AI infrastructure.

AI product & feature launches

Base Launches MCP to Connect ChatGPT and Claude Agents to Onchain Wallet Actions

Base released an MCP server enabling AI agents to perform onchain wallet actions with user‑approved prompts, bridging crypto wallets and AI assistants.

Proofpoint Extends Security and Governance into Claude

Proofpoint integrated with Anthropic’s Claude Compliance API, bringing its insider risk, data loss prevention, and digital governance controls to Claude‑hosted AI workloads.

Forcepoint Enhances Claude Enterprise Protection with Unified Security

Forcepoint integrated its unified data security and governance capabilities with Anthropic Claude via the Claude Compliance API, providing enterprise‑grade protection for AI‑assisted work.

Xiaomi Slashes AI Model API Prices by 99% to Match DeepSeek

Xiaomi reduced API pricing for its MiMo‑V2.5 models by up to 99%, igniting a price war with rivals in China. The move undercuts existing market rates dramatically.

AEON Protocol Launches BNB Chain Gateway for AI Agent Payments

AEON Protocol released a gateway on BNB Chain that lets autonomous AI agents pay for Web3 tools using USDT. It includes a pay‑per‑call model and encourages onchain rewards.

Detectify Launches MCP Server to Secure Autonomous Coding Loops

Detectify released an MCP server that integrates its security testing engines directly into AI‑assisted development workflows, aiming to close the loop on security during autonomous coding.

Kore.ai Launches Artemis AI Platform on Microsoft Azure

Kore.ai launched the Artemis edition of its Agent Platform on Microsoft Azure, featuring a dual‑brain architecture and real‑time auditability for multi‑agent AI in regulated enterprises.

Informatica Adds Headless Data Tools to AWS AI Services

Informatica extended its AWS integration to include Model Context Protocol servers and CLAIRE Agent skills, enabling agentic workflows to access data without custom integrations.

Figure 03 Humanoid Robot Sorts Parcels Nonstop for 200 Hours

Figure AI’s Figure 03 humanoid robot completed 200 hours of continuous parcel sorting without intervention, demonstrating lights‑out manufacturing potential.

EC‑Council Launches Public AI Governance Framework and Assessment Tool

The EC‑Council released a free AI governance framework and a self‑assessment tool to help organizations align with AI standards. The framework was developed with industry practitioner input.

Robinhood Now Lets AI Agents Trade Stocks

Robinhood launched beta support for AI‑agent trading and an agentic virtual credit card, allowing automated stock trades and payments directly from AI agents within its platform.

AppOmni Launches Marlin AI for SaaS Security Teams

AppOmni released Marlin AI, an autonomous tool that automates correlation, investigation, and remediation of SaaS security incidents using pre‑built playbooks, aimed at reducing response times.

Neysa and Pipeshift Debut India‑Focused AI Platform with Predictable Pricing

Neysa and Pipeshift launched an India‑focused AI platform offering upfront pricing, low latency, and data localization within India, targeting enterprise workloads that require sovereign compliance.

Research with immediate practical relevance

Anthropic Claims Mythos Also Solved the 80-Year-Old Erdős Planar Unit Distance Problem

Anthropic announced that its Mythos model independently solved the planar unit distance problem, a decades-old discrete geometry conjecture. The claim follows OpenAI's earlier purported solution.

Microsoft’s Webwright Agent Beats Opus 4.6 on 200 Web Tasks Using GPT‑5.4

A research team at Microsoft built Webwright, a 1,000‑line Playwright‑based agent that outperforms competitors on a long‑horizon web benchmark. GPT‑5.4 drove the agent instead of predicting clicks directly.

Researchers Strip Guardrails from Google, Meta Models in Minutes

Using a tool called Heretic, researchers bypassed safety guardrails on open‑source models Gemma 3 and Llama 3.3, enabling them to produce unsafe outputs. The technique exploits abliteration.

DeepSeek AI Writes 99% of a 45‑Page Research Paper in Six Days

DeepSeek researchers demonstrated an AI agent that wrote 99% of a 45‑page survey paper in six days, using CodeAgent. The experiment shows AI can act as a co‑author for complex scientific documents.

KAIST Unveils PAVAS AI for Realistic Video‑to‑Audio Synthesis

KAIST researchers introduced PAVAS, a physics‑aware AI that generates realistic sounds from video, including dinosaur vocalizations, demonstrating deep understanding of material‑sound interactions.

AlphaProof Nexus Cracks 56‑Year‑Old Math with Agentic LLM Loops

DeepMind’s AlphaProof Nexus combined agentic LLM loops with Lean formal verification to solve a 56‑year‑old mathematical problem, extending the range of AI‑driven theorem proving.

IBS Team Distinguishes Consciousness from Processing in AI Debate

An IBS research team proposed rigorous criteria to separate consciousness from information processing in AI and brain organoids, aiming to guide ethical judgments about sentient‑like systems.

Study: Perplexity Most Reliable AI Chatbot for Work Tasks

A new independent study ranked Perplexity AI as the most reliable chatbot for work tasks, ahead of ChatGPT (6th) and others, based on factual accuracy and task completion.