Telegram bot against CSAM
Protect your community
Elsa globally bans the accounts that distribute child sexual abuse material. AI flags what looks suspicious, but a person always makes the ban decision, and it applies at once in every group of the network.
Official accounts: @elsabanbot and @ggbanbot.
What it bans, and what it doesn't
Elsa bans only for child sexual abuse material. It does not act on spam, insults, scams, arguments or opinions.
The rule we apply
The criterion is article 189.1.d of the Spanish Criminal Code, which since the reform by Organic Law 1/2015 covers virtual child pornography, but requires the material to consist of:
Realistic images of a minor engaged in sexually explicit conduct, or realistic images of a minor's sexual organs, for primarily sexual purposes.
To define what a “realistic image” is, Circular 2/2015 of the Spanish Attorney General's Office concludes:
Therefore, only “realistic images” that closely approximate the graphic representation of an actual minor, or of their sexual organs, can potentially fall within the concept of child pornography. Prosecutors should thus not consider cartoons, manga or similar representations to be included, since they would not properly be “realistic images”, as they do not seek that closeness to reality.
The same Circular notes that “realistic images” “could include altered images of existing people and even computer-generated images”.
Under the principle of unity of action, the Attorney General's circulars bind prosecutors, so this criterion governs the whole public prosecution service in Spain.
Quotations translated from Spanish; the Spanish text is the authoritative one.
In practice
Not banned
Manga, anime or comic-style illustration.
Banned
Hyperrealistic drawing: computer-generated, AI-generated or hand-drawn with such detail that it closely approximates the graphic representation of a real minor.
Banned
Material made by altering the image of an existing person, using the face or body of a real minor.
How it works
Reads the messages
An AI model analyses the text of the messages in the groups where Elsa is present, including photo and video captions.
Compares images
Photos and videos are compared against files a moderator has already flagged.
AI never bans
When a message looks suspicious, at most the sender is muted for one hour to limit the damage, and the case goes to the human moderators.
A person decides
A person always makes the ban decision, and once made it applies in every group of the network at once.
The model is retrained on real messages labelled by hand by a closed group of reviewers, following a shared written criterion.
Get started
- 1
Add Elsa to your group
Tap “Add @elsabanbot” and pick the group you want to protect.
- 2
Make Elsa an admin
With permission to restrict members and delete messages. Without them, Elsa can only watch.
- 3
And forget about it
The rest of the time Elsa stays silent: no replies, no noise. That silence means everything is fine.
Add Elsa only to the groups you want to protect, never to your moderators' chat: there Elsa would act just as in any other group.