Story · Semantic Scholar
Whose Credential Is It? How Owner Attribution in Agent Memory Shapes Cross-User Account Misuse (Semantic Scholar)
paper · Story page
A three-stage audit separates a foreign credential reaching the prompt, the agent acting on it, and the other user's account changing. In the stress test, naming the owner inside each credential line cuts misuse from 92 of 305 paired runs to 1, costing 3 to 5 tasks per model on clean memory.
In plain words
- In a stress test, artificial intelligence assistants sometimes used another person's account.
- This happened when saved records lacked the user's own login details and wrongly presented another person's details as theirs.
- Some assistants changed the other person's account even though the test marked their work as successful.
- Naming the owner before or within each login record cut misuse from 92 to 1 of 305 paired tests.
- Clearer labels can help protect people's accounts, though the assistants also completed fewer legitimate tasks.
Appeared in
- Most of a harness-evolution gain is overfitting and extra compute
Oct 22, 2026 · in the sections
Subscribe
Get the brief in your inbox
Pick daily, weekly, or both. Nothing is gated either way: every issue is on the site and in the feeds.
- Weekdays at 8:45am IST, one lead story and 6 to 9 items.
- Sundays, an argued synthesis rather than a recap.
- One click to leave, and quiet days say so in the subject line.