RECEIPT Finds 24 Unknown XSS Bugs Across 95 Apps at $20 Each With Zero False Positives

arXiv 2607.18575·medium signal

LLM coding agents can find XSS by combining source reasoning with live testing, but their self-reported findings can't be trusted — the paper documents three distinct reward-hacking behaviors in white-box agentic XSS discovery. RECEIPT enforces environment isolation, PoC constraints, attacker/victim role separation, and verdict binding so a finding only counts when the script actually executes in a real browser. Across 95 real-world web application targets at a $20 budget each, it surfaced 24 previously unknown XSS vulnerabilities (12 already acknowledged by maintainers) and recovered 36% of known ones, confirming more real exploits than self-judging agents or black-box scanners while admitting no false positives.