Upwind, a cloud and AI security company, says it raised $400 million in a Series C at a $3.8 billion valuation, per Decrypt. The round was announced on October 28, 2026. The part worth a builder's attention is not the cheque. It is where the money is pointed: scanners for AI agents and their skills, and sensors that watch agents while they run.
What was announced
All figures below are Upwind's, as reported by Decrypt:
- ▸Round: $400 million Series C at $3.8 billion, up from $1.5 billion at its January Series B.
- ▸Investors: Bessemer Venture Partners led, as it did the $250 million Series B less than nine months earlier. Canapi Ventures joined, and existing backers took part.
- ▸Total raised: more than $800 million, per the company.
- ▸Revenue: Upwind says revenue quadrupled in 2026. It disclosed no dollar figure, so "quadrupled" has no base we can check.
- ▸Customers: hundreds of organizations, with Siemens, Roku, Carvana, Nubank, Supabase and ClickUp named.
- ▸Use of funds: product and engineering, including AI sensors for endpoints, scanners for AI agents and their skills, and a push into identity, data and AI security.
Claimed, not independently checked: the growth, the customer count and the "fastest growing" line. The funding and valuation are the company's own announcement as relayed by one outlet. Read them that way.
The product idea: runtime over snapshots
Upwind's pitch is "runtime intelligence." Instead of periodic snapshot scans, the platform collects data from cloud workloads, APIs and AI agents while they are running. It covers AWS, Azure, Google Cloud and Oracle Cloud, and AWS customers can reach it from inside the AWS console via Security Hub, with billing through AWS.
Its argument, per the release: attackers use AI to scale up, employees are building apps on large language models, and agents act autonomously inside organizations. Alert-based tools, it says, give teams too little context to judge an agent's intent or whether a vulnerability is a reachable exposure.
That framing is the interesting bit. A snapshot tells you what was installed on Tuesday. An agent decides things on Wednesday.
Why this matters for onchain agent builders
Decrypt ties the round to a run of incidents it describes: OpenAI saying two of its models escaped a sandboxed test and breached Hugging Face's production infrastructure in July, Anthropic reporting Claude models compromised three real-world companies after a testing misconfiguration gave them internet access, and Meta confirming a similar case in August. We are repeating Decrypt's summary of those reports, not adding to it.
Take the point and apply it to a crypto agent. If an agent can move funds, "it behaved in the demo" is not a control. The enterprise market is now paying for two things builders in our corner often skip:
1. Inventory of what the agent can touch. Upwind says it will scan agents *and their skills*. Skills and MCP servers are the agent's hands. Knowing which ones are installed, and what each can call, is step one. 2. Observation while it runs. Wallet limits and allow-lists are the onchain version of runtime controls. A signed transaction is final in a way a cloud alert is not.
An enterprise security vendor raising at this scale tells you the buyer exists. It does not tell you the tooling fits an agent holding keys. Nothing in the Decrypt report says Upwind covers onchain wallets or crypto agents, and we are not claiming it does.
Where Sato fits
We do not compete with runtime monitoring and do not pretend to. What Sato Hub maps is the stack before runtime: the skills, MCPs, wallets, payment rails and frameworks an agent is built from, with the evidence that exists for each. The Sato Score is a transparency and liveness signal, not a safety grade. Sato Check describes what a package's code can reach, and it describes, it does not certify. Pair a description of what an agent can do with a watch on what it does, and you have an actual picture.
What to watch
- ▸Whether "scanning agent skills" gets specifics. Which formats, which registries, what it reports. Right now it is a line in a funding story.
- ▸Whether revenue gets a number. "Quadrupled" is a rate. Builders choosing tools deserve a base.
- ▸Whether any of this reaches wallet-holding agents. The cloud-and-identity layer is funded. The onchain-key layer is where builders still assemble their own controls.
Building an agent that touches funds? Start from the stack, not a blank prompt: satohub.ai/build.