Lumo 2.0 focuses on privacy: how to assess enterprise AI
Proton introduced Lumo 2.0, an update to its artificial intelligence assistant with an explicit focus on privacy. Gizmodo’s coverage highlights zero-access architecture, infrastructure in Switzerland and a commitment not to use conversations for model training. The release draws attention to a core enterprise issue: performance cannot be the only selection criterion.
What makes the proposition different
Proton says conversations and files are protected so that the company itself cannot read stored content. Lumo 2.0 also adds modes for speed and more complex reasoning, as well as image analysis, web search and team-oriented plans.
These statements matter, but buyers should verify them in contracts, technical documentation and independent evidence. The Gizmodo article appeared in its deals section, another reason to distinguish a product description from a security assessment.
Privacy is not a single setting
A tool may exclude prompts from training while retaining operational logs, metadata, identifiers or temporary copies. Organizations also need to know which subprocessors are involved, where data travels, how long it is retained and how it is deleted.
The assessment must align with applicable privacy laws, customer contracts and internal rules. Sensitive information, trade secrets and third-party data require specific controls and, in some cases, should remain outside the tool.
- Verify encryption, key management and recovery procedures.
- Confirm retention, deletion, export and model-training practices.
- Review corporate identity, multifactor authentication, roles and audit logs.
- Test quality, citations, context limits and language performance.
Security and usefulness must coexist
A private tool that cannot support the work will be bypassed by employees. A highly capable assistant without adequate controls increases leakage risk. Evaluation should combine protection, accuracy, integration, user experience and cost.
A pilot can use synthetic or preclassified data and a representative task set. Subject-matter experts can review outputs while IT evaluates administration, logging and incident handling.
Avoid the false choice between innovation and control
Products that compete on privacy expand the market’s options, but no brand eliminates the need for policy, training and human review. Governance remains the responsibility of the organization using the service.
Conclusion
Lumo 2.0 helps move privacy to the center of enterprise AI discussions. The sound decision is a documented assessment that compares promises with actual controls and tests the tool in the company’s environment. Secure innovation starts with evidence.
How CSP can help
CSP can help compare AI assistants, review privacy and security, classify approved data and run controlled tests with real users. Contact CSP to verify that a solution meets technical, legal and operational requirements before wider adoption.
