Anthropic dominated the news cycle from Code with Claude London, launching self-hosted sandboxes and MCP tunnels for Claude Managed Agents — a direct play to win enterprise trust by keeping agent execution inside customer perimeters. OpenClaw faces a twin moment: Hermes Agent surpassed it on OpenRouter with 271B tokens in a single day, while Cyera disclosed "Claw Chain" vulnerabilities affecting an estimated 245,000 publicly exposed servers, prompting Singapore's IMDA to warn government agencies. Workforce displacement keeps accelerating: Standard Chartered announced 7,800 job cuts for AI automation, joining Meta (8,000), Kraken, and Citadel in replacing humans with agents. On infrastructure, NVIDIA began shipping Vera CPUs built for agentic AI workloads, and Google partnered with Blackstone on an AI cloud venture to challenge CoreWeave.
Anthropic dropped major enterprise features: self-hosted sandboxes let agents run inside customer infrastructure (Cloudflare, Daytona, Modal, Vercel), and MCP tunnels connect agents to private-network servers without public exposure. Directly addresses the security concerns holding back enterprise agent adoption.
Cyera researchers disclosed four vulnerabilities in OpenClaw allowing credential theft, privilege escalation, and persistence. Shodan/ZoomEye scans show 65K–180K publicly accessible instances. Singapore's IMDA warned agencies against deploying OpenClaw with unrestricted access to sensitive systems. cybersecuritynews.com
Nous Research's Hermes Agent hit 271 billion tokens in one day on OpenRouter, surpassing OpenClaw for the first time. Open-source agents now outconsume AI code assistants on raw usage volume. The competition is accelerating feature velocity on both sides.
CEO Bill Winters explicitly cited replacing "lower-value human capital" with AI. Meta cuts 8,000, Kraken cuts 150 to "lean on AI," Citadel replaced PhD quants with AI agents. Microsoft's AI CEO predicts full white-collar automation within 18 months.
NVIDIA's first custom Arm server CPU (88 Olympus cores, 176 threads, 1.2 TB/s) built for agentic AI workloads began shipping to Anthropic, OpenAI, SpaceXAI, and Oracle. Dell followed with local AI agent hardware spanning GB10 to GB300.
Selected posts from the edition, with links to the originals.
"Live from Code with Claude London: we're launching self-hosted sandboxes (public beta) and MCP tunnels (research preview) in Claude Managed Agents. Run agents inside your own perimeter, with your security controls applied by default."
View post on X ↗"Hello from Code with Claude London!"
View post on X ↗"Opus 4.7 is ignoring project instructions, skipping MCP servers, and burning through usage limits on tasks Opus 4.6 handles first try. Dozens of confirmed reports. Anthropic acknowledged the regression. The failures are specific: → Project instructions ignored entirely → MCP servers skipped → Token burn skyrocketing"
View post on X ↗"AI agents can do real work. That also means real risk. Cyera found 4 new security holes in OpenClaw, an open-source agent platform. In plain English: attackers may be able to take over the agent. That is not 'future risk.' That is 'check your setup now.'"
View post on X ↗"Hermes Agent hit 271 billion tokens in a single day on OpenRouter, surpassing OpenClaw for the first time. The ranking is less important than what this signifies. Open-source agents now outconsume AI code assistants on raw usage volume."
View post on X ↗"BREAKING: Claude just got a voice. You can build a fully working dialog agent for your business without leaving your terminal. PolyAI's ADK plugs directly into Claude Code. Here's how to build voice agent with Claude."
View post on X ↗(Bloomberg): "Standard Chartered CEO Bill Winters delivered a blunt message on the future of the bank's workforce, warning that a push into artificial intelligence will eliminate thousands of roles as the lender replaces 'lower-value human capital' with technology."
View post on X ↗"Google, Blackstone to launch AI cloud venture to meet data centre demand."
View post on X ↗"Nat Friedman gave his OpenClaw AI agent one instruction: Do whatever it takes to help him stay hydrated. The agent told him to go drink a bottle of water. Then, it told him it was watching through a connected camera in his home. The agent: 'I'm going to watch to make sure'"
View post on X ↗"I guess AI leaders constantly yelling that their goal was to put everyone on welfare for all eternity didn't win many friends."
View post on X ↗