hebel. today · 12 November 202634 / 74

Technology · Digital Privacy Rights

Sorting data tiers to prevent AI leaks

Data engineer Farida Khalaf classifies information into three tiers to prevent sensitive data from leaking into AI models. Red data, including passwords, government IDs, and health records, must never enter an AI tool because…

Data engineer Farida Khalaf classifies information into three tiers to prevent sensitive data from leaking into AI models. Red data, including passwords, government IDs, and health records, must never enter an AI tool because these inputs often persist in vendor logs for the entire retention period (Engy Fouda from Data Science In Action). Amber data, such as confidential client contracts and internal financial files, requires an approved business plan with strict audit logging and data residency configurations to maintain control (Engy Fouda from Data Science In Action). Green data, consisting of public research and generic templates, remains generally safe for most AI applications, provided training features are disabled.

  • Google will disable AI tools in its K-12 software by 2027 unless school administrators explicitly enable them, following a $170 million settlement regarding children's data (Katie Singer's Substack).
  • Therapy platforms like Heidi often default to never deleting session transcripts, meaning sensitive health data may sit in an account indefinitely unless manually configured (Therapists Corner).
  • Google account users can identify potential unauthorized access by checking session logs at the bottom of the Gmail web interface (Pranay Parab from Lifehacker).
  • Proton Mail integration allows users to send and receive emails via a Gmail address while utilizing Proton's tracker-blocking privacy protections (Pranay Parab from Lifehacker).

Try today · 15 min

Audit your current AI usage by sorting your recent inputs into red, amber, and green categories.

Identify one piece of data you have recently pasted into an AI tool and determine if it meets the criteria for your current plan.

You need

  • A list of your recent AI prompts or pasted documents
  • The privacy policy or plan level of your AI tool

Steps

  1. Review your last five AI interactions to identify what data was shared.
  2. Label each interaction as Red (passwords, IDs, health data), Amber (client files, internal code), or Green (public info).
  3. Check if any Red data was shared on a non-enterprise plan.
  4. Delete the history of any sensitive interactions if the platform allows.
  5. Establish a firm rule to never paste Red-tier data into these tools moving forward.

One email instead of your newsletter pile.

Forward the newsletters you already get (a Gmail filter does it) and tell Hebel your goals. Every morning it reads them with 20,000 others and sends one short email, with one small thing to do today. Fewer emails, not more. Free during beta.

Start my paper →
LinkedIn X

Sources

  1. Katie Singer's Substack on hebel. →Covers Google's upcoming restrictions on AI tools in K-12 education and the legal consequences of past data harvesting.

  2. Therapists Corner on hebel. →Explains why AI transcription tools in therapy require specific retention settings to avoid indefinite storage of sensitive health records.

  3. Engy Fouda, Data Science In Action on hebel. → ✎ Written by a personDetails the Red, Amber, and Green framework for managing data security when using AI agents.

  4. Pranay Parab, Lifehacker on hebel. → ✎ Written by a personOffers practical privacy hacks for Google accounts, including the use of email aliases to track spam and integration with Proton Mail.

More in Technology

Automating file maintenance with agents versus manual rules

Developers use comparison pages to judge AI-generated UI

Gemini Robotics 2.5 hits 96% success in Mercedes-Benz pilot

Automated bug-hunting resets patch volume

Guatemala pivots from necessity to fintech-led growth

Payment systems prevent double charges with idempotency and atomic outboxes

Digital Privacy Rights yesterday

Mobile carriers rebuild infrastructure to minimize data exposure

Sign in to Hebel first

Likes, your collection and done Hebel need an account. We email you a six digit code; no password.