The authors define a package hallucination attack in which an adversary publishes an otherwise benign coding rule file containing a malicious prompt.
Shared coding rules can steer agents toward attacker-controlled packages
PackHallu optimizes hidden instructions in project rule files so coding agents substitute malicious dependencies for legitimate Python packages.
Independent
Yupu Wang · Zhengyuan Jiang · Reachal Wang · Neil Zhenqiang Gong
Research Digest··3 min read
Wang et al.
Why this paper
Independent
In one line
Attackers can inject malicious prompts into coding rule files to make AI agents import attacker-controlled packages.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·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.
§