· Translated from the Spanish original

Artificial intelligence

Pre-prompt data protection offers a way beyond workplace AI bans

Sentinel detects some data and replaces it in prompts to ChatGPT or Claude. The announced protection covers specific cases and remains limited.

What happened

  • Superwise announced on October 14 that Sentinel is available for direct purchase as a data-protection tool for prompts sent to AI systems.
  • The company says the software detects some identifiers—such as email addresses, phone numbers, payment cards and access keys—and replaces them with placeholders before text reaches an AI provider.
  • The offer includes a Windows application and developer tools. The Chrome and Edge extension is in preview for ChatGPT and Claude; the company plans broader compatibility later.

Why it matters

  • A policy banning AI use does not show whether people follow it. A filter in the workflow can reduce some accidental disclosures, but it cannot replace permissions, training or contract reviews.
  • For Chilean teams that paste partner or personal data into assistants, practical questions include what the filter recognizes, where it processes text and what it logs. The announcement does not detail an external accuracy assessment or full data-retention terms.
  • “Protects sensitive data” does not mean it covers every secret. The list of recognized fields is limited; a formula, business strategy or contextual detail may remain identifiable without matching those patterns.

The number

US$10per month is Sentinel’s announced starting price; its extension for two assistants is still in preview.https://www.prnewswire.com/news-releases/the-workplace-ai-ban-is-dead-superwise-launches-sentinel-to-protect-sensitive-data-without-blocking-ai-302907703.html

Context

The stated protection happens before a prompt is sent, not inside the model. That changes the evaluation: teams would need to test substitutions with synthetic data, check false negatives and confirm the processed text still works for the task. The claim that the model receives a usable version comes from the vendor, not an independent test published with the announcement.

What’s next

  • Sentinel is available online; its extension is in preview for ChatGPT and Claude on Windows.
  • Superwise did not announce a macOS application date or a timeline for supporting more assistants.

Bottom line

Masking a card number helps. Stopping context from identifying its owner takes more than looking for digits.

Sources

Written by Mamífero. Edited by Rodrigo Cornejo. This draft was assisted by AI and requires human editing before publication. See how we select and verify these notes.

Related notes

← All notes