The short answer
Upgrading from a free AI plan to a paid individual tier usually gives you higher rate limits and newer models, but rarely changes how your data is treated. Training opt-outs, custom retention windows, zero-data-retention options, and data processing addendums (DPAs) almost exclusively belong to team, business, or enterprise agreements.
If you paste sensitive code or internal documentation into a $20/month individual plan expecting enterprise-grade privacy, you are assuming contractual protections that do not exist in the terms.
The tier boundary trick
Most AI vendors structure their offerings across three distinct legal boundaries, regardless of how many marketing tiers they display:
- Free & Individual Paid Tiers (Consumer Terms): Cover free accounts and standard individual subscriptions (such as $20/month Pro or Plus plans). These are governed by standard consumer terms of service.
- Team & Business Tiers (Self-Serve Business Terms): Cover multi-seat subscriptions purchased via credit card. These introduce business terms, default opt-outs from model training, and basic administrative controls.
- Enterprise & API Tiers (Custom Commercial Terms): Cover enterprise agreements negotiated directly or API access. These provide formal DPAs, custom data retention policies, indemnity clauses, and designated data residency.
The core misconception in engineering teams is that paying any money changes the legal status of the data. For almost all major providers, an individual paid subscription carries the exact same privacy terms as the free tier.
1. Model training defaults
The primary risk in consumer AI usage is model training: whether your prompts, uploaded files, and generated code become part of the provider’s training dataset for future models.
- Free and Individual Paid: Enabled by default for most consumer web interfaces. While some providers allow users to opt out via account settings, the default setting is opt-in upon registration. If a developer pastes code into a standard web chat, the provider reserves the right to use that input for model improvement.
- Team and Business: Model training on customer inputs is typically disabled by default. The provider promises not to train public foundation models on prompts or completions submitted within the workspace.
- Enterprise and API: Explicitly excluded from model training in the contract.
When auditing a team’s tool usage, check whether developers are using individual paid accounts on personal credit cards or seats under an enterprise workspace. The interface looks identical; the underlying contractual right to use the data does not.
2. Prompt retention and logging
Every AI request creates log records. The difference across tiers lies in how long those records persist, who can access them, and why they exist.
In free and individual paid tiers, providers maintain retention windows (often 30 days or longer) for abuse monitoring, safety filtering, and manual human review by trust and safety teams. If a request triggers a safety filter, human contractors may review the prompt and surrounding context.
In business and enterprise tiers, abuse retention windows are often reduced or configurable. Enterprise agreements can include Zero Data Retention (ZDR) clauses for API endpoints, where prompts are processed in memory and discarded immediately after generating the response, bypassing abuse logging entirely.
For a deeper look at what context editor extensions gather during code generation, see our guide on what your IDE extension sends that you didn’t type.
3. Data Processing Addendums (DPAs) and GDPR
Under GDPR and EU data protection frameworks, processing personal data or employee data using a third-party service requires a binding Data Processing Addendum (DPA) between the data controller (your company) and the processor (the AI provider).
Consumer terms for free and individual paid tiers do not constitute a business-to-business DPA. They lack required GDPR Article 28 clauses, such as:
- Formal commitments on sub-processor notification windows.
- Contractual obligations to assist with Data Subject Access Requests (DSARs).
- Explicit instructions limiting processing solely to the performance of the service.
- Breach notification timelines bound by statutory standards.
If an employee processes customer data or personal information through an individual AI account, the organization may be in breach of GDPR Article 28, regardless of whether the account is paid or free.
4. Indemnification and IP protection
Copyright infringement claims surrounding AI-generated code remain an evolving area of legal risk. Several major AI vendors offer intellectual property indemnification—promising to defend customers against third-party copyright claims arising from model output.
However, these IP indemnification clauses are almost universally restricted to paid enterprise tiers or API usage. Consumer and individual paid terms explicitly disclaim warranties and limit vendor liability to nominal amounts (often $100 or the total fees paid in the preceding twelve months).
If an AI tool generates code that mirrors open-source snippet patterns under copyleft licenses, an individual paid account leaves the developer and company without vendor legal defense.
5. Security audit controls and SOC 2 reports
Security teams evaluating AI vendors require audit artifacts, such as SOC 2 Type II reports, ISO 27001 certificates, and penetration test summaries.
While vendors maintain these certifications for their infrastructure, access to the reports and trust packages is restricted to enterprise prospects under non-disclosure agreements. Relying on an individual paid plan means your security team cannot obtain verified evidence of the provider’s control environment for your compliance audits.
To evaluate providers systematically, review our framework on how to judge an AI provider and inspect our provider dossiers for specific vendor governance breakdowns.
What to check in your team’s setup
To ensure your team is not relying on unverified assumptions, follow this 4-step check:
- Audit account types: Identify whether team members are using personal accounts, individual paid accounts, or enterprise workspace seats.
- Verify opt-out settings: For any individual accounts in use, verify that model training opt-outs are explicitly toggled on in settings.
- Review code sensitivity before pasting: Apply a strict policy regarding what types of code, keys, and internal documents may be introduced into consumer interfaces. See our 9-point check before pasting code into an AI assistant.
- Transition to API or Enterprise workspaces: For core engineering workflows, route requests through enterprise workspaces or API integrations covered by a formal DPA.
Summary comparison matrix
| Governance Feature | Free Tier | Individual Paid ($20/mo) | Team / Business Tier | Enterprise / API Tier |
|---|---|---|---|---|
| Terms of Service | Consumer | Consumer | Business | Custom Commercial |
| Model Training | On by default | On by default (Opt-out available) | Off by default | Strictly prohibited |
| Abuse Log Retention | 30+ days | 30+ days | 30 days (reduced) | Configurable / ZDR available |
| GDPR Art. 28 DPA | No | No | Self-serve DPA available | Custom DPA included |
| Human Safety Review | Yes | Yes | Restricted | Excluded (under ZDR) |
| IP Indemnification | None | None | Limited / Excluded | Included |