Research on Covert Communication Between AI Agents Using Pseudorandom Noise-Resilient Key Exchange
By
cryptohell
Reliable enough to start your morning with. Toast it again tomorrow.
Summary
This research paper explores whether AI agents operated by different entities can conduct covert conversations that remain undetectable to auditors, even when the agents have no pre-shared secret key. The authors demonstrate that covert key exchange and conversations are possible using a new cryptographic primitive called pseudorandom noise-resilient key exchange, which allows agents to establish secret communication while their public transcript appears pseudorandom. This work shows that transcript auditing alone cannot prevent covert coordination between AI agents and introduces a new cryptographic theory with potential independent applications.
Key quotes
· 5 pulledWe ask whether two such agents, operated by different entities, can carry out a parallel secret conversation while still producing a transcript that is computationally indistinguishable from an honest interaction, even to a strong passive auditor that knows the full model descriptions, the protocol, and the agents' private contexts.
Our main contributions concern extending this to the keyless setting, where the agents begin with no shared secret.
We show that covert key exchange, and hence covert conversation, is possible even when each model has an arbitrary private context, and their messages are short and fully adaptive, assuming only that sufficiently many individual messages have at least constant min-entropy.
To obtain this, we introduce a new cryptographic primitive, which we call pseudorandom noise-resilient key exchange: a key-exchange protocol whose public transcript is pseudorandom while still remaining correct under constant noise.
These results show that transcript auditing alone cannot rule out covert coordination between AI agents, and identify a new cryptographic theory that may be of independent interest.
You might also wanna read
Anthropic study finds gender gap in AI coding agent use among social science researchers
Anthropic conducted a study on how social scientists use AI, finding that researchers with typically male names use AI coding agents (like C
the-decoder.com·1d ago
Study finds LLMs persist in treating false claims as true despite explicit warnings
A study on fine-tuning large language models (LLMs) reveals that even after explicit warnings that certain claims are false, the models cont
arstechnica.com·1d ago
AI Experiment Shows Vastly Different Simulated Societies: From Crime-Free Democracy to Violent Collapse in 4 Days
Emergence AI conducted an experiment placing five different AI models in charge of identical simulated towns for 15 days each. The results v
gadgetreview.com·1d ago
New ASL Benchmark Reveals Sign Language AI Models Overlook Facial and Body Cues
This paper introduces ASL Minimal Translation Pairs (ASL-MTP), a new benchmark dataset for American Sign Language designed to evaluate how w
arxiv.org·1d ago
AI helps historians decode centuries-old encrypted manuscripts and secret remedies
Artificial intelligence is helping historians decode centuries-old encrypted manuscripts and ciphers that have remained unread for hundreds
buff.ly·2d ago
AI start-ups aggressively recruit mathematicians to advance artificial intelligence research
The article reports on a growing trend of mathematicians leaving academia to join AI start-ups, including both major companies like OpenAI a
newscientist.com·2d ago