Is It Safe to Use AI for Sales?

The answer fits in one line. AI helps your sales, but the prospect must not end up in the prompt. A sales team pastes sensitive data every day. A contact's name. Their email and phone. A deal's value. Call notes. In a consumer AI, this data can be retained, reviewed or reused for training. And you are the one deciding on that use. Under the GDPR and the CCPA, that choice is your responsibility, not just the vendor's. There is a clean method: anonymise the identifiers and the deal details before you send.

What a sales team really pastes into AI

The use cases are everywhere in a sales cycle. Draft an outreach email. Summarise a customer call. Score a lead. Prep a follow-up. Each time, the temptation is the same. You copy the CRM export or the transcript, and you paste the whole thing into the AI.

The problem is not the AI. It is what you feed it. A CRM export holds real personal data. So does a call transcript. The AI does not need those values to work. It needs the structure of the message.

GDPR: you are the controller

The GDPR (Regulation (EU) 2016/679) sets a simple rule. Whoever decides why and how personal data is processed is the 'controller'. When a rep pastes a name, an email or notes into the AI, it is their company that processes that data. Not the AI vendor alone.

Article 5 sets the accountability principle. You must be able to demonstrate that the processing follows the data-protection principles. Article 24 adds a concrete duty. You must put appropriate technical and organisational measures in place.

Article 5(1)(c) carries another key principle: data minimisation. Data must be adequate, relevant and limited to what is necessary. Sending a full CRM record to draft an email is hard to justify. The model only needs the shape of the message. That is the technical basis for anonymising before the prompt.

CCPA: prospect data fits the definition

In the United States, the California Consumer Privacy Act (CCPA) defines 'personal information' broadly. It is any data that identifies a consumer or household, or can reasonably be linked to one. Names, emails, purchase records, inferences: it all counts. The prospect data pasted into AI tools falls squarely within this definition.

The CCPA is enforced by the California Privacy Protection Agency and the California Attorney General. Civil penalties can reach $2,500 per violation. They rise to $7,500 per intentional violation. Covered businesses must implement reasonable security practices. They must also have contracts in place with third parties that receive personal information.

OWASP: an AI is not a vault

OWASP lists Sensitive Information Disclosure as the LLM02 risk in its 2025 Top 10 for LLM applications. An AI system can unintentionally expose private or confidential data through its outputs. That exposure can come from data placed in the prompt, held in memory, or embedded in training data.

OWASP describes LLMs as probabilistic text generators, not secure vaults. In rare cases, a model exposed to raw CRM data can surface one prospect's details to another user. This is not a routine event, but a real risk. And even accidental disclosure can leave the organisation responsible for missing safeguards.

The fix: anonymise, contract, keep a human

Good news: AI is still an excellent sales copilot. It can draft, summarise and score without ever seeing the real values. The principle is simple. You replace each identifier and deal detail with a token. The AI reasons about the structure. You restore the real values afterwards, locally.

Anonymising is not a guarantee of compliance. It is risk reduction and a real minimisation lever. Pair it with a few simple habits, summarised below.

1. 1Anonymise prospect identifiers and deal details before the prompt.
2. 2Prefer a CRM-integrated tool with a signed DPA (Data Processing Agreement) and a no-training clause.
3. 3Check the product's current terms: consumer tiers may retain and review data differently from enterprise tiers.
4. 4Respect call-recording consent according to your jurisdiction.
5. 5Keep a human check on every output before it reaches the prospect.

• What the AI should see: the email frame, the objection type, the summary structure.
• What the AI does not need to see: the name, email, phone, deal value, or target account.

That's what ONYRI Sanitize is for. The engine detects sensitive data — name, email, phone, deal value, notes — and replaces it with reversible tokens before sending. Detection and the mapping stay in your browser. Only anonymized text reaches the model. The AI finds only tokens, never your real prospects. You gain the speed of AI, without handing a third party data the prospect never gave you for that.