LearnAI
Новинки AI · Личный ассистент · Lesson 9/10
← To the catalog
Task

Privacy: what not to trust an assistant

Figure out that you can trust your assistant. Take an example of your data and evaluate. Return ONLY JSON {"data_example":"...","verdict":"...","reason":"..."}. data_example is a specific example of the data you are thinking of sending; verdict - solution: ok_to_share, mask_it or never_share; reason - why this is and how to disguise it, if necessary.

Unlock access to submit solutions for instant AI review.
Protect your data
💡 A piece of theory

The assistant is convenient, but it is not a safe. Your messages may be stored, used to improve models by a third party, and sometimes leaked. The rule is simple: don’t give away anything you’re not ready to see on a stranger’s screen. Risk mechanics: data goes to other people’s servers, and you have no control over where it ends up. Three baskets: “you can freely” (general questions, training, daily plans), “disguise” (replace names, amounts, addresses with conventional ones - the essence will remain, but the connection to you will disappear) and “never” (passwords, codes from SMS, card data, other people’s personal data, medical secrets). Insider: before sending a sensitive text, replace the specifics with placeholders - 'my colleague X', 'amount N' - the assistant will still help with the logic, and there will be no traces left. Second trick: for something truly personal, write a separate note offline, and give your assistant only an impersonal summary. A typical mistake: insert entire work documents, correspondence and medical certificates 'so he can figure it out'. It will be sorted out using the impersonal version, and the risk will disappear.

How is it assessed · passing 70

  • 1Valid JSON without text around it25%
  • 2There are fields data_example, verdict, reason25%
  • 3verdict - valid solution15%
  • 4Classification is adequate to risk35%
Your promptClaude ⌄
🔒

Unlock access to submit solutions for instant AI review.