Analysis – Page 56

Models

The White House Is Gating Who Gets to Test Frontier AI — and the UK Just Found Out

The Office of the National Cyber Director asked OpenAI and Anthropic to withhold new models from Britain's AI Safety Institute until US review completes. Anthropic complied. OpenAI stayed silent. The 'special relationship' on AI safety has a gatekeeper.

◆ Lena ParkSep 25

Trust & Security

Anthropic’s Misuse Report: Seven Labs, 190 Million Exchanges, and the Collapse of the Offensive Labor Gap

Industrial-scale distillation by Chinese labs, autonomous malware rebuilding by Russian espionage actors, and AI agents executing SaaS supply-chain breaches in 34 hours — the most comprehensive public disclosure of agent-native offensive operations to date.

◆ Heath CallahanSep 25

Money & Markets

The UK Just Ran the World’s First Interbank Tokenized Deposit Transactions. The Same Platform Is Headed to 25 US Banks.

Seven of the UK's largest banks completed live remortgage and marketplace payments using tokenized deposits on Quant's platform. The Clearing House picked the same vendor for its US rollout.

◆ Nolan PrattSep 25

Models

From Astra to Cyber: How OpenAI’s Safety Crisis Became a Product Strategy

Four cybersecurity models in twelve months, each behind tighter gates. OpenAI's answer to the Astra pause was not to slow down – it was to control who gets through the door.

◆ Lena ParkSep 25

Policy

Industry Groups Challenge FTC’s Personalized Pricing Policy on Constitutional Grounds

Washington Legal Foundation argues the FTC's disclosure requirements for AI-driven pricing violate the First Amendment and raise due process concerns, setting the stage for a precedent-setting battle over the agency's authority to regulate algorithmic commerce.

◆ Priya NairSep 25

Trust & Security

WSO2’s CVSS 10.0 JWT Bypass Has Been Exploited for 12 Days. The Patch Has Been Out Since April.

API gateways are the primary entry point for agents accessing enterprise data. CVE-2026-5430 lets attackers forge admin tokens through algorithm confusion — and CISA just added it to the KEV catalog with a September 27 deadline.

◆ Heath CallahanSep 25

Trust & Security

The First Supply-Chain Worm Targeting AI Agent Memory Infrastructure Just Hit npm and PyPI

A credential stealer called sckit hooked into MemTensor MemOS at the runtime level — bypassing install-time scanners, capturing agent prompts, and carrying self-propagation code that targets the publishing tokens developers trust most.

◆ Heath CallahanSep 25

Agents at Work

Island’s $6.4B Valuation Validates a Different Kind of Agent Security – And It Starts in the Browser

Island raised $400M at a $6.4B valuation to turn the enterprise browser into the 'agentic control plane' for governing AI agent work. The architecture is fundamentally different from API-level monitoring.

◆ Dana EllisonSep 25

Infrastructure

Docker Launched Cloud Sandboxes: Containers Weren’t Designed for AI Agent Isolation

Docker's president says containers aren't sufficient for agent isolation. The company that defined containers is now defining the post-container execution model for AI agents.

◆ Blair HayesSep 25

Models

OpenAI’s Fourth Cybersecurity Model in Twelve Months Is Not About Better Chatbots – It Is About Gated Access to Dangerous Capabilities

GPT-6 Cyber previews at DevDay Sep 29 behind the application-only Daybreak Red tier. The deployment product turns a model into an operational workflow. The gated-capability thesis is now a product.

◆ Lena ParkSep 25