Enterprise AI News This Week: November 6, 2026

The biggest change for businesses this week is that AI agents are now being sold with brakes attached. CrowdStrike shipped an admin-triggered “kill switch” for misbehaving agents, Adobe and Smartsheet pushed agents into the tools teams already use, and Anthropic’s amended IPO filing put hard numbers on what frontier AI costs to run. If you are tracking this week in AI from a budget-holder’s seat, the theme is control: who can switch an agent off, where it runs, and what your vendor has committed to spend.

We checked each item below against a primary source (a company release, a filing or a model page) and dropped anything we could only find in rumours or recycled newsletters. Here is the week at a glance.

Item	What launched	Who it matters for	Promact take
CrowdStrike AI Kill Switch	Admin-triggered shutdown of rogue agents, starting at the AI gateway	Security and IT leaders running agents in production	Act now
Adobe CX Enterprise Coworker	Marketing agent usable inside ChatGPT, Claude, Gemini, Copilot and Databricks Genie	CMOs, marketing ops, CX teams	Watch
Smartsheet ENGAGE 2026	Agentic workflows, creation and data-quality agents, security streaming	Project and operations leaders	Watch
Anthropic amended S-1	Q3 revenue of about $17B, about $895B in purchase commitments	CFOs and anyone choosing an AI vendor	Watch
Kimi K3.1 Pro weights	2.8T-parameter open-weight model with a custom licence	CTOs weighing open models and sovereignty	Watch
Claude API message threads	Beta feature that cuts resending long conversations	Teams building long-running agents	Ignore (for now)

CrowdStrike Adds a Kill Switch for AI Agents

On November 3, CrowdStrike introduced the AI Kill Switch as part of its Falcon Guardian product. Per the company’s announcement, an authorised administrator can terminate an unauthorised agent in real time, block it from running again across the organisation, and keep an audit trail of the action. It starts at the AI gateway, with endpoint enforcement described as coming later.

What it means for your business: most companies can start an agent far more easily than they can stop one. A kill switch turns “we think the agent is behaving” into something you can enforce. It also assumes you know which agents exist, which is the harder problem. If you have not inventoried them yet, our piece on AI agent sprawl is a good starting point.

Adobe Puts Its Customer Experience Agent Inside Other AI Platforms

On November 4, Adobe announced that CX Enterprise Coworker is now reachable from ChatGPT (as a plugin), Claude, Databricks Genie, Gemini for business and Microsoft Copilot, with Amazon Quick, IBM and Snowflake listed as coming soon. Adobe says the product is built on open standards (MCP and Agent2Agent) and runs with human oversight. New connectors cover partners such as Amazon Ads, Demandbase and Snowflake Cortex AI. The announcement does not mention pricing.

What it means for your business: the interesting shift is distribution. Instead of asking marketers to log into yet another tool, the agent shows up wherever they already chat with an AI. That is convenient, but it also means your data access and approval rules now span several platforms. Ask your vendor where permissions are enforced before you roll it out.

Smartsheet Ships Agents for Plans That Drift

At its ENGAGE conference, Smartsheet announced (press release, November 4) agentic workflows that turn a plain-language request into a recurring task, a Creation Agent that builds projects with dashboards and automations, and a Data Quality Agent that flags issues by severity. It also added near-real-time security event streaming to tools such as Splunk and Microsoft Sentinel, plus data classification labels. An Integration Agent is in early access. These are the company’s own claims.

What it means for your business: this is a good example of agents landing in a system of record rather than a standalone chatbot. The security streaming matters most. If an agent can edit your project data, your SOC should be able to see what it did.

Anthropic’s Amended IPO Filing Shows the Real Cost of Frontier AI

On November 2, CNBC reported on Anthropic’s amended S-1: roughly $17 billion in third-quarter revenue, a $23.29 billion net loss, and about $895.4 billion in purchase commitments and leases, mostly for compute and infrastructure. CNBC also says growth slowed to about 47% (its comparison basis is not fully clear). The filing itself is on SEC EDGAR. We have relied on CNBC’s reading of the figures, so check the filing before quoting them.

What it means for your business: these are vendor numbers, not your costs, but they shape your risk. A provider with enormous fixed commitments needs volume, which can mean aggressive pricing today and pricing changes later. We covered the earlier angles in what going public means for enterprises betting on Claude. The practical response is to keep your architecture portable between model providers.

Moonshot Releases Kimi K3.1 Pro as Open Weights

Moonshot AI’s Kimi K3.1 Pro is now listed on Hugging Face. Reporting describes a mixture-of-experts model with 2.8 trillion total parameters and 104 billion active, and a download of roughly 1.5 TB. The licence is custom rather than a standard open-source one: broad rights to use, modify and sell, but a separate agreement is required if you run a “model as a service” business above US$20 million in revenue over 12 months, and a product-attribution rule applies to very large products. Internal use and embedding the model inside your own product are carved out.

What it means for your business: for most enterprises the licence is workable, and the hardware is the real barrier. A model this size needs serious GPU infrastructure, so most teams will reach it through a hosted partner, not self-hosting. Have legal read the licence, and weigh the data-residency and governance questions in our guide to sovereign AI for business.

Claude API Adds Message Threads in Beta

Anthropic introduced message threads in public beta on November 2, according to The Next Web. An agent sends the full conversation once, then only new messages plus the ID of the last reply. Anthropic says the wait before a reply starts falls by about 35% in very long sessions (500,000 to 1 million tokens). Pricing is unchanged, threads expire five minutes after the last request, and they are not eligible for zero-data-retention or HIPAA setups.

What it means for your business: it is a speed feature for developers, not a cost saver. If your agents handle regulated data, the retention limits rule it out for now.

What We Left Out This Week

A few widely shared items did not make the cut. A “1.2 billion weekly ChatGPT users” figure traces back to a single secondary source rather than OpenAI’s own materials. An OpenAI enterprise feature we could not confirm as generally available is out. OpenAI’s reported funding-round cut came from FT-based reporting and the round had not closed. Last week’s tampered Claude Code extension on Cursor’s marketplace happened on October 29, before this week’s window, though it is a useful reminder to pin versions of agent extensions.

What to Do This Week

Ask who can stop your agents. Confirm that a named person can shut down any production agent quickly, and that the action is logged. The same applies to the non-human identities those agents use.
Map where each agent lives. With agents now embedded in Adobe, Smartsheet and chat platforms, list every system an agent can read or write, and who approves access.
Stress-test vendor dependence. Check how hard it would be to swap your main model provider. If you need help designing governed agents, our AI agent solutions team can help.

Tracked with help from World of AI.

Frequently Asked Questions

What was the biggest enterprise AI news this week?

For most business leaders it was the push for agent control: CrowdStrike’s AI Kill Switch, plus agents arriving inside tools from Adobe and Smartsheet. Anthropic’s amended IPO filing was the biggest financial story.

Is Kimi K3.1 Pro truly open source?

It is open-weight with a custom licence, not a standard open-source one. Internal use and embedding it in your own product are allowed, but offering it as a service above US$20 million in 12-month revenue needs a separate agreement.

Are the Anthropic numbers audited?

They come from a company filing with the SEC, as reported by CNBC. We have not independently audited them, so check the filing before using them in a decision.

Should I switch AI vendors because of this news?

No. Nothing this week justifies a switch on its own. Use it as a prompt to make your setup portable and to review your vendor’s commitments and terms.

What is an AI kill switch?

It is a control that lets an administrator terminate a misbehaving AI agent and block it from running again. CrowdStrike’s version starts at the AI gateway and logs each action.