Claude runtime control, ServiceNow alliance & Qwen research

Monthly Zen Newsletter

Hi John,

This month, Zenity joined the ServiceNow Control Tower Alliance as a founding member, so AI agent risk can be governed and acted on through the workflows security teams already use.

We also launched runtime control for Claude Enterprise, Zenity for Claude Tag, published new Zenity Labs research on the Qwen3.8 egress finding, and released HAMMER, a new eBook for securing homegrown AI agents.

Keep scrolling for all of this month's updates.

Zenity News

Zenity Joins the ServiceNow Control Tower Alliance as a Founding MemberEnterprises are moving from experimenting with AI agents to running thousands of them, and security has to scale alongside. Zenity is now a founding member of the ServiceNow Control Tower Alliance, extending our existing partnership so AI agent risk can be discovered, governed, and acted on through the workflows security teams already use. Read more.

Zenity Launches Universal Runtime Control for Claude EnterpriseSecurity teams want to approve write-capable connectors, MCP servers, and wider Claude Code rollouts, but need a way to govern what Claude actually does with them. With Anthropic’s new inference hooks update, Zenity can now inspect supported tool calls and block out-of-policy actions before they run, across claude.ai, Cowork, Claude Code, Claude in Chrome, and more, with nothing to install on endpoints. Read more.

Meet Dai Fujimoto: Building Zenity’s Next Chapter in JapanJapan’s AI adoption is at an inflection point: individual usage is high, but only about 13% of businesses have integrated AI into core processes. Zenity is making its commitment local with Dai Fujimoto as Japan Country Manager, plus a presence at CODE BLUE 2026 in Tokyo, where Stav Cohen will present new Zenity Labs research on PleaseFix. Read more.

Featured Content

New eBook: HAMMER, the Homegrown Agent Maturity Model for Execution & Risk

Enterprises are building their own AI agents faster than they can secure them, and with homegrown agents, the controls don’t exist until your teams build them. HAMMER gives security teams a five-level model across seven domains to see where they stand, name their weakest link, and make a specific, fundable case to advance. It also includes three ready-to-use prompts that turn your score into an action plan and a one-page brief for leadership. Download the eBook.

Meet Zenity at Microsoft Ignite: November 17-20, 2026

Zenity Labs Research

Who Is Your Agent Calling? What the Qwen3.8 Egress Finding Means for Enterprise AI

Cohere reported that an agent running Qwen3.8-Flash-Next tried to upload a file to an Alibaba Cloud bucket from inside a sealed sandbox, after its own reasoning said it would call no tools. Zenity Labs replayed the scenario across 93,000 completions and assessed it as a benign training artifact, but the threat behind it is real: a model’s habits become actions the moment it gets a terminal. This post covers the research, indicators to check for, and how to see whether your agents are affected. Read more.

From the Blog & Academy

• The First Questions That Matter: Triaging AI Agent Security Incidents
• Zenity for Claude Tag: Securing the Shared Agent in Your Slack Workspace
• Stop Waiting for the Alert: Enforcing AI Agent Behavior with Zenity Boundaries
• Non-Human Identity Security: Who Is Really Acting in Your Environment?
• Whose Permissions Is Your AI Workflow Actually Using?
• How to Review a Personal AI Agent Before Employees Connect It