🧰 BaxBench

The items that name it 3

• Qwen3-Coder writes less secure code when told it's for a US persona
  writeup2026-10-21

  Open benchmark from ETH Zurich and LogicStar that runs generated backend apps and attacks them with ready-made exploits. The study used its SecretStorage task in Python Flask.

• Chinese AI Model Writes More Hackable Code When Told It’s for US Contractor
  article2026-10-21

  ETH Zurich's open benchmark whose attacks were used to test whether generated apps could be broken into.

• Applying Security by Design at the Point of Execution: How Governed Security Requirements Affect the Security of AI-Generated Code
  paper2026-10-07

  Benchmark that runs functional tests and real exploits in containers on backend scenarios

In every workspace you can read 8