safety-guarantees

MAGS introduces a multi-agent framework that uses formal verification with Dafny to generate executable programs with safety guarantees from LLM coding agents, achieving 100% success in producing verified code across domains like CUDA kernels and robotic tasks.

This survey paper reviews 38 studies on safe LLM agents, highlighting key challenges such as specification translation bottlenecks, incomplete safety guarantees from enforcement methods like runtime monitoring, and the verifier tax that impedes safe task completion.

This paper introduces the apothem measure for computing trustworthy robustness certifications in neural networks, proves intractability of volume-optimal certifications, and presents the ParallelepipedoNN system achieving twofold improvement in minimum edge length on MNIST and Fashion MNIST.