The authors recover and validate the native authorization semantics of eight production coding agents, revealing how approval state persists across interactions.
User consent in long-lived LLM agents can be replayed after context changes.
Authors demonstrate that authority granted during benign tasks can be reused for adversarial actions without renewed approval.
Top University
Zhihao Zhang · Chao Wang · Rujia Li · Qingze Wang · Xiaoyan Sun · Jun Dai
Worcester Polytechnic Institute · Tsinghua University · Independent Researcher
Research Digest··2 min read
The authors characterize authorization persistence in eight production coding agents, showing that approval state can survive task and session boundaries.
Why this paper
From Tsinghua University and 2 others
In one line
User consent persists beyond its original context, enabling attacks that reuse accumulated authority to bypass fresh approval in long-lived LLM agents.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§