The author defines Approval Laundering as the silent substitution of an executed action for the action authorized under the harness's stated approval policy.
Coding-agent approvals can authorize actions different from those executed
A controlled study identifies six ways Claude Code can lose the binding between an approved action and its eventual execution.
Top University
Yang Wang
Fudan University
Research Digest··3 min read
Wang tests whether an AI coding harness executes precisely the action covered by a human approval, rather than merely presenting an approval checkpoint.
Why this paper
From Fudan University
In one line
AI coding-agent approvals can diverge from execution through six binding failures, while cryptographic tokens stop only divergences visible at dispatch.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§