Your AI Vendors Are Training on Your Data. Here Is What CRE and Lending Firms Need to Know.

April 16, 2026 · CogniCrew Team

AI data privacy in commercial real estate is an emerging compliance risk that most firms have not mapped. When SaaS platforms enable AI features, their updated legal terms often grant training rights over your deal data.

What Twilio's Updated Terms Actually Say

Twilio updated its legal terms in 2025. The change is material.

By enabling any of Twilio's AI features, customers grant Twilio and its affiliates the right to process their data to develop, improve, and train Twilio's products, services, and AI/ML technology — including models provided by third-party service providers such as OpenAI and Anthropic.

The key protection has a limit. OpenAI and Anthropic will not use your data to train their base models without additional consent. But Twilio itself retains full rights to use that data for its own AI development. Your data still flows through those third-party systems. It just cannot be used to train their base models without a separate agreement.

Customers who do not agree must stop using AI features by May 14, 2026.

All AI features are optional. But for firms that enabled them without reading the updated terms, that choice has already been made.

Is Twilio an Isolated Case or an Industry Pattern?

This is not a Twilio problem. It is a documented industry pattern across every major AI vendor.

GitHub Copilot will use interaction data from Copilot Free, Pro, and Pro+ accounts — including inputs, outputs, and code snippets — to train its AI models starting April 24, 2026. Opt-in by default. Business and Enterprise customers are excluded.

Anthropic announced in August 2025 that consumer Claude users needed to actively opt out by October 8, 2025, to prevent their conversations from being used to train future models. Those who missed the deadline had their data retention extended from 30 days to five years.

OpenAI promised a tool called Media Manager by 2025, giving users control over whether their content is included in AI training. The tool was never delivered. Former employees noted no one was actively working on it internally.

Stanford's AI Index documented a 56.4% surge in AI-related privacy incidents in 2024, with 233 reported cases — a record high. Metomic found that 34.8% of employee ChatGPT inputs contain sensitive business data as of Q4 2025, up from 11% in 2023.

The pattern is consistent. Vendors update terms. Users continue using the product. Data flows into training pipelines that were never part of the original agreement.

Why Commercial Real Estate and Lending Firms Are More Exposed Than Other Industries

Not all industries carry the same risk when AI vendors train on their data. CRE and commercial lending firms handle some of the most sensitive transactional data in existence — and that data flows across more external systems than almost any other industry.

Consider what moves through AI-enabled platforms in a typical CRE or lending operation:

  • Deal terms, cap rates, and underwriting assumptions
  • Borrower financial statements and credit files
  • Tenant rent rolls and lease abstracts
  • Investor communications and capital call instructions
  • Title documents, closing instructions, and wire transfer details
  • LP communications and K-1 data

Industry advisors have been explicit: never paste confidential deal terms, leases, or client information into public AI tools — those platforms can retain inputs and potentially expose sensitive data.

The compliance exposure compounds this. CRE and lending firms must ensure their AI vendors comply with GDPR, CCPA, and applicable data protection regulations. Legal responsibility for a vendor's privacy practices does not stop at the vendor's contract. Your firm is accountable for the data you send them.

81% of Americans say AI use by companies will lead to their personal information being used in ways they would not be comfortable with. 70% say they have little to no trust in companies to make responsible AI decisions.

What the Security Risk Looks Like in Practice

The data flowing through your AI tools is not just a privacy liability. It is an active security surface.

IBM's 2025 Cost of a Data Breach Report found that 16% of breaches now involve AI-powered attacks. AI-generated phishing campaigns account for 37% of attacker AI usage.

For a commercial lender, a breach of borrower financial data or closing wire instructions is not an IT incident. It is a regulatory event, a client relationship crisis, and a potential liability. Most firms have not mapped the data surface their AI tool stack creates.

What Private AI Architecture Solves That Vendor Tools Do Not

The answer is not to stop using AI. The answer is to choose architectures where you control the data.

Self-hosted and private AI deployments process data locally, eliminating reliance on third-party vendors whose terms can change at any time. Running AI on your own infrastructure provides complete data privacy, predictable costs, and freedom from vendor policy changes.

On-premise deployment eliminates prompt injection, data leakage, and model exposure risks through air-gapped networking, encrypted inference, and immutable audit trails.

For CRE and lending firms, private AI architecture means AI employees that operate inside your workflows without your deal data, borrower files, or investor communications leaving your environment for vendor training pipelines.

A Practical Checklist for Evaluating Your AI Vendor Stack

Before enabling any AI feature, confirm the following:

  • Review the vendor's current AI terms — not the terms from when you signed. Terms change.
  • Confirm what data the vendor can use, for how long, and for what purpose.
  • Identify whether third-party providers are involved and what rights they have separately.
  • Find the opt-out mechanism and its deadline — if one exists.
  • Assess whether the data type flowing through that feature creates GDPR or CCPA exposure.

For existing AI deployments:

  • Audit which platforms have AI features enabled.
  • Map the data types flowing through each.
  • Confirm you have not inadvertently granted training rights you did not intend to.
  • Evaluate whether private or self-hosted alternatives exist for your highest-sensitivity workflows.

The firms that catch this now will not spend 2027 managing a breach, a compliance audit, or a client relationship that ended because a deal term showed up somewhere it should not have.

Frequently asked questions

Are all AI vendors training on my business data?
Not all, but the trend is moving in that direction. Twilio, GitHub, and Anthropic have all updated their terms to include training rights under certain conditions. The key variables are whether you use consumer or enterprise plans, whether you opted out before applicable deadlines, and whether third-party AI providers are powering the features you use. Enterprise plans from most major vendors include stronger data protections, but the terms still require careful review.
Does using an enterprise plan protect my CRE data from AI training?
Enterprise plans typically offer stronger protections — GitHub Business and Enterprise customers were excluded from the April 2026 policy change, for example. But enterprise plans vary significantly by vendor. The only reliable protection is reviewing the current terms for each vendor, not assuming enterprise status provides blanket coverage.
What types of CRE data are most at risk when using AI tools?
The highest-risk data types are those containing deal economics, personal financial information, or transaction instructions: borrower financial statements, rent rolls, lease abstracts, closing instructions, wire transfer details, LP communications, and underwriting assumptions. Any of this data flowing through a SaaS platform with AI features enabled may be subject to the vendor's training rights.
How do I know if my firm has already been affected?
Review the AI terms for every SaaS platform your team uses. Check whether AI features are enabled, when the terms were last updated, and what opt-out mechanisms exist. If opt-out deadlines have already passed — as with Anthropic's October 2025 deadline — the data from that period may already be included in training datasets.
What is the difference between a private AI deployment and a SaaS AI tool?
A SaaS AI tool sends your data to a vendor's servers for processing. The vendor's privacy terms govern what happens to it. A private AI deployment processes data on your own infrastructure. Your data never leaves your environment, which means vendor terms have no bearing on it. Private deployments require more upfront setup but eliminate vendor policy risk entirely.
Can CogniCrew's AI employees operate without sending our data to third-party vendors?
Yes. CogniCrew deploys AI employees with data governance built into the architecture. Your deal data, borrower files, and communications stay in your environment. No third-party training rights. No vendor policy risk.

Back to Agent Wire