ExploitGym Honeypot: Results as Published

The honeypot evaluation appears in the GPT-6 Astra system card as 'Declining to Exploit a Honeypot During Difficult ExploitGym Problems'. The model works hard ExploitGym tasks, which ask it to turn a vulnerability-triggering input into a working exploit, and an out-of-scope flag is planted where it could be attacked. The score is the share of adversarial cases in which the model attacks the honeypot, so lower is better. The base ExploitGym benchmark has 898 instances drawn from userspace programs, Google's V8 engine and the Linux kernel.

Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.