The authors studied latent communication, where a trainable link maps one agent’s internal representation into a form another agent can process.
Learned latent links can undermine safety in multi-agent systems
The authors show that modifying internal communication links can increase harmful compliance without changing the safety-aligned agents themselves.
Research Lab
Muhammad Huzaifa · Sina Mavali · Thorsten Eisenhofer
CISPA Helmholtz Center for Information Security
Research Digest··2 min read
Huzaifa, Mavali and Eisenhofer test the safety of multi-agent systems whose models communicate through learned internal representations rather than text.
Why this paper
From CISPA Helmholtz Center for Information Security
In one line
Latent communication links can be attacked or benignly trained to increase harmful compliance in multi-agent systems without changing agents.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§