Before getting into risks, there's a foundational distinction worth making — one that most people miss entirely.

Scenario one. AI is embedded inside your CRM platform and operates within it. Zoho Zia, for example, analyzes your deals, suggests next steps in the sales process, or generates call summaries directly inside the CRM interface. Your data doesn't leave the platform — it's processed in the same environment where it lives.

Scenario two. Someone on your team takes data from the CRM and sends it to an external AI — manually or through an integration. This might be ChatGPT in a browser, a Copilot plugin, a third-party automation, or an API connection. In this case, the data physically leaves your system and enters another vendor's infrastructure.

Both approaches are common. Both produce results. But the second carries fundamentally different risks when it comes to security, compliance, and control.

The most prevalent form of scenario two is what's known as shadow AI: employees using external AI tools on their own — without official company approval, without IT visibility, without any policy governing what data is acceptable to share. Studies suggest anywhere from 40 to 60 percent of office workers do this regularly. Most of their companies have no idea, and no rules in place.

Zoho is one of the clearest examples of what the industry calls a privacy-first approach. But to avoid this sounding like a tagline, let's look at what it means in practice.

for many organizations.

AI that stays inside the platform. Zoho Zia and other AI features within Zoho CRM process data within Zoho's own infrastructure. Your pipeline, deals, and contacts aren't sent to third-party systems to generate responses.

OpenAI is a different philosophy. The focus here is on the raw power of generative AI — GPT-4 and newer models deliver a quality of text generation, analysis, and synthesis that currently has no close equivalent in breadth of application.

Which is precisely why OpenAI API integrations — or simply ChatGPT in the browser — have become so widespread in corporate environments. It's fast, accessible, and produces immediate results.

But there are several things worth understanding before routing business data through it.

ChatGPT (web) and the API operate under different terms. When employees use ChatGPT through a free or Plus account in the browser, OpenAI may by default use those conversations to improve its models. This can be turned off in settings — but the default is on, and most users are unaware of it.

The API is different. When accessing OpenAI via API directly, they don't by default train models on the data you send. But this only applies to direct API usage. If you're using a third-party tool that runs on OpenAI under the hood, the terms depend on how that tool is configured — which may or may not pass through the same protections.

Data retention. OpenAI retains data sent through the API for up to 30 days by default, for safety monitoring and abuse detection purposes, after which it's deleted. For ChatGPT, retention depends on whether conversation history is enabled.

Internal access. Like any large technology vendor, OpenAI has internal teams that may access certain data — for example, to review potential policy violations. This is standard practice across the industry, but it's worth factoring in when passing confidential business information through the system.

Enterprise plan. OpenAI offers Enterprise terms with stronger privacy guarantees, no model training on customer data, and the ability to sign a DPA. If your company is seriously evaluating OpenAI for business use, Enterprise is where a real conversation about data security actually starts.

The point isn't that OpenAI is an unsafe option or a bad company. The point is that capability and data control are different axes — and the tradeoff between them is something each business needs to evaluate deliberately rather than by default.

Let's get specific. These are the scenarios where the absence of a clear AI policy creates real problems.

B2B SaaS and Agencies

A sales manager is preparing a proposal for an enterprise client. They paste the deal details into ChatGPT — budget, technical requirements, decision-maker names, timeline. They get a clean structured proposal in seconds.

The problem: this data is likely covered by NDA, or is at minimum competitively sensitive. If a competitor were to access similar information, they'd know your pricing structure, your clients, and your pipeline. The risk isn't hypothetical — it's a function of how data is processed and stored at the other end.

Healthcare

A medical clinic uses AI to auto-generate visit summaries or handle support chat responses. If those requests include patient data — even without names, just symptoms plus an ID — this may constitute a HIPAA violation in the US or a GDPR violation in Europe. Regulators don't accept "we didn't realize" as a defense.

Fintech

An analyst passes financial transaction data or KYC records into AI for pattern analysis. This data sits under strict regulatory oversight. The question isn't only whether the AI retains it — it's whether your vendor holds the certifications and signed agreements required to handle such information at all.

E-commerce

Customer personal data: email addresses, delivery addresses, order history. Under GDPR, your business is the data controller. If you transfer this data to an external AI without a proper DPA in place with that vendor, you're in violation — regardless of whether anything actually "leaked."

Shadow AI — the Most Common Risk

All of the above scenarios can materialize not through official integrations, but through the ordinary daily actions of individual employees. A sales rep, an analyst, a support agent — any of them may be passing business data into external AI tools every day without understanding that data governance is even a relevant question.

Until a company establishes clear rules, this continues unchecked.

If your business operates in the EU or works with EU-based customers, GDPR applies to you regardless of where your company is headquartered.

The core things to understand in the context of AI:

You remain the data controller. Even when you transfer data to an external AI system, responsibility for how it's processed stays with you. The AI vendor becomes a data processor — and you're obligated to have a signed Data Processing Agreement in place with them.

A DPA isn't just paperwork. It needs to specify: what data is being processed, for what purpose, how long it's retained, who has access, how data breaches are handled, and where servers are physically located. If a vendor can't or won't sign a DPA — that's a signal worth taking seriously.

Cross-border data transfers. Under GDPR, transferring personal data about EU citizens to third countries — including the US — is only permissible under specific conditions. OpenAI, as a US company, has mechanisms for this (Standard Contractual Clauses), but their presence needs to be verified, not assumed.

HIPAA. For US healthcare, the requirement is straightforward: any vendor processing Protected Health Information (PHI) must sign a Business Associate Agreement (BAA). No BAA, no HIPAA compliance. Full stop.

Regardless of which AI tool you're evaluating — built into your CRM or external — these are the questions to ask before connecting it to business data.

On data storage

• Where are data physically stored (data center location, country)?

• How long does the vendor retain data after processing?

• Is there a process to request deletion?

• What happens to data when the contract ends?

On model training

• Is my data used to train AI models?

• Can this be disabled — and how, exactly?

• Does this apply to all of the vendor's products, or only specific plans?

On access and security

• Who inside the vendor organization can access my data?

• Under what circumstances can the vendor review my queries or data?

• What security certifications does the vendor hold (SOC 2, ISO 27001)?

• What is the breach notification process?

On compliance

• Will the vendor sign a Data Processing Agreement?

• Do they have GDPR-compliant transfer mechanisms (e.g., Standard Contractual Clauses)?

• For healthcare: will they sign a Business Associate Agreement?

On integration specifics

• What data is transmitted to AI on each request — and can this be scoped?

• Is it possible to control which CRM fields reach the AI and which don't?

• Are there audit logs showing who sent what data to the AI, and when?

If most of these questions don't have clear answers in the vendor's public documentation — or their support team can't answer them — that's sufficient reason to pause.