Can I Put Company Data into ChatGPT or Claude? A Legal & Compliance Guide for Swiss SMEs (April 2026)

The short answer first: A Swiss SME may in principle put company data into ChatGPT and Claude, but only in the commercial tiers, with a signed data processing agreement (DPA), EU/Switzerland data residency enabled and a documented risk assessment. The free and personal paid versions (ChatGPT Free/Plus/Pro/Go, Claude Free/Pro/Max) have been effectively off-limits for business personal data since the Anthropic change of 28 August 2025.

This guide summarises the legal landscape under Swiss revDSG, EU GDPR and the EU AI Act as of April 2026, and gives you a clear decision matrix.

Swiss law: revDSG applies technology-neutrally to every AI input

The revised Swiss Data Protection Act (revDSG, in force since 1 September 2023) is technology-neutral and directly applicable to any AI-driven data processing. The Swiss data protection authority EDÖB confirmed this on 8 May 2025 and clarified at Data Protection Day 2026 that the DSG applies fully as soon as personal data is processed in generative AI (EDÖB update, 24.09.2025).

Three provisions are operationally decisive for your SME:

• Art. 5 revDSG, Controllership. When an employee enters personal data into ChatGPT or Claude, your company is the controller for that processing.
• Art. 9 revDSG, Processor relationship. Without a written DPA, the AI provider acts as an independent controller and may reuse your data under its own rules.
• Art. 16–18 revDSG, Cross-border transfers. Since 15 September 2024 the US qualifies as adequate under the Swiss-U.S. Data Privacy Framework (DPF). Both OpenAI and Anthropic are DPF-certified.

For professional-secrecy holders (Swiss Criminal Code Art. 321) and banks (Banking Act Art. 47), the bar is higher: disclosure to an external provider is only permissible if it is strictly necessary and reasonably foreseeable for the client. Trade secrets are additionally protected under Art. 162 PC and Art. 6 UCA, pasting a pricing model into a consumer chatbot can trigger criminal and unfair-competition liability.

The Anthropic turning point of 28 August 2025: Claude Free/Pro/Max is now off-limits

The single most important change of the reporting period: Anthropic announced on 28 August 2025 that chats and coding sessions from Claude Free, Pro and Max are used by default for model training from 8 October 2025 onwards, unless the user disables the "Help improve Claude" setting. Retention jumps from 30 days to 5 years (Anthropic, 28.08.2025).

The pre-enabled toggle placement qualifies as a "deceptive design pattern" under EDPB guidelines, casting serious doubt on the validity of that consent under GDPR (The Decoder, 08/2025).

Crucially, this does not apply to commercial products: Claude for Work (Team/Enterprise), the API, AWS Bedrock, Google Vertex AI, Claude Gov and Claude for Education are never used for training. Companies on the Team plan (from 20 seats) are protected.

The tier matrix: which product may see which data?

This decision matrix links product tier and data category (as of April 2026):

• ChatGPT Free / Plus / Pro / Go, NOT SUITABLE for company data. Training on by default, no DPA, no residency. Use only for public research and brainstorming without personal data.
• Claude Free / Pro / Max, NOT SUITABLE for company data. Training on by default since 8 October 2025, 5-year retention. Same restrictions as consumer ChatGPT.
• ChatGPT Business (ex-Team, 25 USD/user). No training, DPA available, but no EU/CH data residency as of April 2026. Fits low- to medium-sensitivity internal data.
• ChatGPT Enterprise, first choice for personal data. No training, DPA, Europe Data + Inference Residency available (EEA incl. Switzerland), BAA via sales. Available since February 2025.
• Claude Team / Enterprise (from 25 USD/user). No training, DPA. But no native EU/CH residency. For GDPR/revDSG-compliant processing from Switzerland, the mandatory path is Claude via AWS Bedrock eu-central-1 (Frankfurt) / eu-central-2 (Zurich) or Google Vertex AI europe-west3.
• Azure OpenAI in Switzerland North, optimal Swiss sovereignty. Microsoft DPA, data at rest in Switzerland, MAM/ZDR and BAA available. Recommended for sensitive personal data and trade secrets.
• OpenAI API with ZDR (Zero Data Retention). Zero retention, DPA, EU residency. For the highest sensitivity, pricing models, strategic documents, professional-secrecy data.

The real threat landscape: what SMEs learned in 2025–2026

Shadow AI is measurable. Check Point Research reported for October 2025 that 1 in 44 GenAI prompts in enterprise environments contains highly sensitive data, and that 87 percent of organisations with regular GenAI usage are affected. By November 2025 the share climbed to 22 percent sensitive prompts across an average of 11 GenAI tools used in parallel per company (Check Point Research).

Concrete incidents in the reporting period:

• ChatGPT share-link indexing (July/August 2025): Over 100,000 private conversations became Google-searchable until OpenAI disabled the feature on 1 August 2025.
• Mixpanel incident (November 2025): Attackers breached OpenAI sub-processor Mixpanel via smishing; names, emails, approximate location and org IDs of API users were exposed (OpenAI Incident Report).
• ShadowLeak (September 2025): The first zero-click server-side vulnerability in the ChatGPT Deep Research agent.
• Malicious Chrome extensions (Dec 2025 / Jan 2026): More than 900,000 and 8 million installs respectively intercepted ChatGPT, Claude, Gemini and Copilot conversations, one even carried Google's "Featured" badge.
• Claude Code leak (31 March 2026): Anthropic accidentally published ~513,000 lines of unencrypted source code via an npm source map.

OpenAI's own CISO Dane Stuckey admits: "Prompt injection, much like scams and social engineering on the web, is unlikely to ever be fully 'solved'" (OpenAI, 22.12.2025).

EU AI Act: Art. 4 AI Literacy applies to Swiss SMEs too

The EU AI Act reaches Swiss providers whose output is used inside the EU (Art. 2(1)(c)). Since 2 February 2025, Art. 4 (AI Literacy) obliges providers and deployers to ensure sufficient AI competence among staff and contractors, regardless of risk class or company size. A documented yearly training is therefore mandatory for your SME as well.

Three non-negotiable rules for your SME

• No consumer tiers for business data. ChatGPT Free/Plus/Pro/Go and Claude Free/Pro/Max must not be used for company data, the single largest shadow-AI risk today.
• DPA before go-live. The OpenAI DPA (effective 1 January 2026) and Anthropic Commercial Terms with integrated DPA (since January 2026) must be signed and on file.
• Document the legal basis. A written balancing-of-interests test under Art. 31(1) revDSG, and a Transfer Impact Assessment where EU data is involved, are mandatory.

Bottom line: safe AI is a question of the right layers

The question "Can I put company data into ChatGPT or Claude?" has a clear, differentiated answer in April 2026: yes, but only in the right tiers with the right protective layers. The investment in enterprise plans (around CHF 25–30 per employee per month) is not a luxury, it is insurance against double-digit fines under Art. 60 revDSG and criminal exposure under Art. 162 / 321 of the Swiss Criminal Code.

But the biggest open flank remains shadow AI. Even with enterprise licences, employees keep using private ChatGPT Plus subscriptions, personal Claude accounts and unknown Chrome extensions on the side. Policies alone don't solve this, you need a technical governance layer that sees what data is actually leaving the company.

This is exactly where hegyon.ai fits in. Our Swiss-built personal AI assistant sits between your employees, your data and the LLM providers. Two guards form the governance layer: the Sensitive Data Guard prevents ultra-sensitive personal data, trade secrets or pricing models from ever reaching an external LLM. The Compliance Guard enforces your internal policies in real time, from signature authority limits to mandatory DPIAs. Swiss hosting, full audit trails, revDSG- and EU-AI-Act-ready. Your employees work productively with modern AI, and you keep control over what leaves your company.