πŸ•΅οΈ ANONYMOUS SURVEY

Shadow Model Census β€” Ai Agents 2026

"What models is your company actually running?" Vendor press releases say one thing. Procurement dashboards say another. This anonymous census asks the people who see production traffic every day β€” no names, no emails, no company names, no tracking. The aggregate becomes the public census.

What is a shadow model?

A shadow model is any AI model running inside a company without official approval or tracking β€” adopted through APIs, side projects, or personal accounts. Each one carries unvetted data-handling, retention, and deprecation risk.

Honest banner: this form collects no identifying fields β€” no name, no email, no company name, no IP logging. Submissions are stored only in your browser's localStorage as a device-only demo so you can see how the aggregate view works. The published census launches after enough responses come in β€” subscribe to the newsletter to get it the day it drops.

What the official surveys already tell us

Which LLM do most AI agents 2026 teams actually run?

GoodFirms' AI-tool tracking of 14,022 active companies: ChatGPT β€” 11,359 companies Β· Claude β€” 10,103 Β· Gemini β€” 4,294. At the developer level it's neck and neck: Stack Overflow's 2026 survey found Anthropic 74.4% vs OpenAI 74.1% usage β€” but 49% of developers want to use Anthropic next year vs 32.1% for OpenAI.

The shadow gap

Stack Overflow's 2026 developer survey found that only 24% of companies have published an approved AI toolset or a formal AI policy β€” leaving the majority of developers making tool choices individually. If you run a model at work that procurement doesn't know about, congratulations: you're shadow AI, and this census is for you.

Enterprise frontier-LLM spend

A June 2026 research compilation (McKinsey, BCG, Gartner, Deloitte, IBM, KPMG surveys) estimates ~85–90% of large enterprises have at least one production LLM deployment, 55–65% run multiple frontier LLMs concurrently, and enterprise frontier-LLM API spend topped $15B in H1 2026 β€” on track to exceed $35B for the full year. Estimated enterprise spend share: OpenAI ~40–45%, Anthropic ~25–30%, Google ~15–18%, Meta 5–8%.

πŸ“‹ The census β€” 3 questions, 30 seconds, zero identity

1. Which models is your company actually running? (check all that apply)
2. Main use case
3. Company size band

No name, no email, no company name, no IP. Stored only on this device for this demo.

πŸ“Š Live aggregate β€” this device (demo)

0 responses on this device

FROM INGENIOUS AI

ModelMinder is our model deprecation radar: it tracks the models in your stack β€” pricing moves, version retirements, quiet downgrades β€” so procurement always knows what's actually running and what it's about to cost.

Our platforms are being upgraded and return November 1 β€” get early access now, or get the whole suite:

Get ModelMinder early access β†’
Get early access on WhatsApp →
See the full Security Suite β†’
"What models is your company actually running?" β€” anonymous 3-question census. No names, no emails, no tracking.
ingeniousai-security-checks-mirror Β· Shadow Model Census
Share on XShare on LinkedIn
NEWSLETTER

The published census launches after enough responses β€” get the results (and every research drop) in your inbox.

Subscribe to the newsletter β†’

More from Ingenious AI

FAQ

Is the Shadow Model Census anonymous?

Yes. The form asks for no names, no emails, no company names and collects no IP addresses. Responses are stored only in your browser's localStorage as a device-only demo so you can see how the aggregate view works.

Why is it called a "shadow" model census?

Because much enterprise AI use is shadow IT: Stack Overflow's 2026 developer survey found only 24% of companies have published an approved AI toolset or formal AI policy, leaving the majority of developers making tool choices individually.

Which LLM do most companies use?

GoodFirms' AI-tool tracking of 14,022 active companies finds ChatGPT used by 11,359 companies, Claude by 10,103, and Gemini by 4,294. Stack Overflow's 2026 developer survey found Anthropic and OpenAI usage neck and neck at roughly 75% each.

How much are enterprises spending on LLMs?

A June 2026 compilation of major research-firm surveys estimates enterprise frontier-LLM API spend topped $15B in H1 2026, on track to exceed $35B for the full year β€” with 85–90% of large enterprises reporting at least one production LLM deployment.

When will the published census be released?

After enough responses come in, the aggregated results will be published as a public census. Subscribe to the Ingenious AI newsletter to be notified the day it launches.

Disclaimer: survey statistics cited above come from the linked third-party research (GoodFirms via ssquares.tech, Stack Overflow 2026 Developer Survey, presenc.ai research compilation) and are reproduced as published β€” methodologies and sample populations differ. This demo form stores responses only on your device.

Frequently asked questions

Is the Shadow Model Census anonymous?

Yes. The form asks for no names, no emails, no company names and collects no IP addresses. Responses are stored only in your browser's localStorage as a device-only demo.

Why is it called a "shadow" model census?

Because much enterprise AI use is shadow IT: Stack Overflow's 2026 developer survey found only 24% of companies have published an approved AI toolset or formal AI policy.

Which LLM do most companies use?

GoodFirms' AI-tool tracking of 14,022 active companies finds ChatGPT used by 11,359 companies, Claude by 10,103, and Gemini by 4,294. Stack Overflow's 2026 developer survey found Anthropic and OpenAI usage neck and neck at roughly 75% each.

How much are enterprises spending on LLMs?

A June 2026 compilation of major research-firm surveys estimates enterprise frontier-LLM API spend topped $15B in H1 2026, on track to exceed $35B for the full year.

When will the published census be released?

After enough responses come in, the aggregated results will be published as a public census. Subscribe to the Ingenious AI newsletter to be notified the day it launches.

Related checks