Securing Web Agents with the Compile-Then-Execute Paradigm

Evidence record

Read the original source ↗

Claim-level citations (3)

• supportsBaya compiles a web-agent plan before exposing it to live pages: In the authors' tests Baya scored 0% attack success on WASP versus 16.2% for API ReAct and 25.5% for browser ReAct; utility was 77.8% over 581 valid WebArena Verified tasks.

  arXiv abstract: 581 WebArena tasks, 77.8% utility; WASP attack-success results

• supportsBaya compiles a web-agent plan before exposing it to live pages: The Baya paper describes compiling the user request to a Python plan using a trusted website action API before observing any live website data.

  arXiv abstract: compile-then-execute paradigm and Trusted Website API

• supportsBaya compiles a web-agent plan before exposing it to live pages: Baya reported 77.8% aggregate utility on 581 valid WebArena Verified tasks, against 66.7% for API-enabled ReAct and 55.1% for browser-only ReAct in the authors' evaluation.

  Abstract, WebArena Verified results and ReAct baselines