Measuring LLMs’ ability to develop exploits

We've developed two new, challenging academic benchmarks measuring AI models’ ability to develop exploits, and an updated version of the benchmark measuring smart contract exploitation.