[pmontra]

> Security by design: In order to protect young people from grooming, internet services and apps shall be secure by design and default. It must be possible to block and report other users. Only at the request of the user should he or she be publicly addressable and see messages or pictures of other users. Users should be asked for confirmation before sending contact details or nude pictures. Potential perpetrators and victims should be warned where appropriate, for example if they try to search for abuse material using certain search words. Public chats at high risk of grooming are to be moderated.

The "publicly addressable" part is feasible and already implemented by many messengers.

Asking for "confirmation before sending contact details" is feasible but easily circumvented as is my mail is joe at example dot com.

Asking for "confirmation before sending [...] nude pictures" is where it gets interesting. How without scanning every file one is about to share?

[9dev]

> Asking for "confirmation before sending contact details" is feasible but easily circumvented as is my mail is joe at example dot com.

I’m pretty confident this is a task where LLMs will shine.

[raxxorraxor]

Not if you want to scan the image. Locally that would still need quite a lot of processing power.

[besse]

I don't understand. Why would anyone care about circumventing a "Are you sure?" dialog?

[pmontra]

There are many ways to share contact details. One is to use an app provided mechanism, which can ask confirmation about sending and receiving contacts. Another one is to just type them into a message. The confirmation of the intention to send is implicit but no confirmation is possible at the receiving end: they read the message and get the contact, willing or unwilling. All they can do is delete it before remembering it and report the other party. There is a provision about reporting unwanted interactions.

About circumvention, one could think about a way to detect contact details in messages. There are several problems with that.

1. If it warns the sender, the sender will try another way until the obfuscated contact passes unnoticed (text, image, vocal, link, whatever.)

2. If it automatically reports the sender there will be many false positives, some due to copy and paste, keyboard autocompletion, etc.

3. We're back to automatic spying of people.