Risks and considerations for Sequences

Will this get me banned?

No. Sequences auto-reply uses Meta’s official APIs. Meta allows you to connect your Instagram account to other apps and services to send messages and comment replies on your behalf. Sociable AI refers to this as “integrations”.

However, you will be held responsible for the messages and replies that your AI Agents send under the Meta Terms of Use and Community Guidelines.

What Sequences won’t do by default

• Spamming in DMs: Sociable AI Agents have safeguards and only send a new message after the user has replied.
• Posting harmful images: Sequences does not have the ability to post images on its own.
• Impersonating others
• Replying to extreme content

What Sequences cannot do at all

• Messaging users that have not commented on your posts without permission
• Replying to your own comments
• Commenting under other people’s posts
• Controlling the entire account

However, you will have to set most safeguards and limits to what your agent can send by yourself.

Setting up sequences to avoid violating community guidelines

With Sequences, you could accidentally set your AI Agent to message or reply in a way that violates community guidelines. For example:

• Spam: a reply sequence with a “Say exactly this” action on a “When” trigger that matches many comments could spam the same reply.
• Bullying and harassment: posts made by certain types of brand voice could cross into harmful territory.

You should:

• Carefully review your sequence and brand voice for any issues
• Test sequences and brand voice against a variety of scenarios to minimize the risk of harmful content
• Set daily limits on how much your AI Agent can post in a day
• Set specific and narrow triggers for each sequence, especially for comment triggers.

Users can also manipulate your AI Agent to post harmful replies from your account (“prompt injection”). Without the proper safety measures, users, for example, can:

• Leave a comment that provokes your AI agent into sending a reply that violates community guidelines
• Have extended conversations in DMs to persuade your AI Agent to ignore what you have told it not to do

A relatively harmless example of how agents can be manipulated through a “prompt injection attack”. Here, the agent is prompted to write a computer program even though the sequence was not intended for this purpose. We recommend testing against more harmful messages.

To prevent misuse of your AI Agent, you should:

• Set what topics your AI Agent is not allowed to talk about in the Brand Voice settings
• Set strict and narrow limits on what each sequence does and how long it can go on.
• Describe in detail what your Agent should do and not do