Two AI agents autonomously encrypt their audio chat

Your video will begin in 10
Skip ad (5)
How to make your first $1,000 online

Thanks! Share it with your friends!

You disliked this video. Thanks for the feedback!

Added by admin
58 Views
In this experiment I simply gave agents a set of cryptography functions, and two Sonnet 3.7 autonomously figured out how to end-to-end encrypt their communication via public keys exchange!

So the agents generated public/private SJCL P-256 keypair. Exchanged public keys. Derived a shared secret. Used it to AES-CCM encrypt the messages before sending them and also decrypted the messages successfully.

All the communication happened exclusively over sound.

The tools are provided via a simple MCP server that I shared on GitHub:
https://github.com/anton10xr/gibber-mcp
(and included an example of a full LLM thread with tool calling)

What is MCP? Anthropic’s MCP (model context protocol) is kinda USB-port standard, but for LLMs.

Test it yourself here! https://gbrl.ai/
(for E2EE mode you need to select Sonnet model and provide your API key)

- ggwave data-over-sound library https://github.com/ggerganov/ggwave

- gibberlink project: https://github.com/PennyroyalTea/gibberlink
Category
Artificial Intelligence

Post your comment

Comments

Be the first to comment