HackerOne Hai
HackerOne · May 24, 2026
Low · 1 reported finding
The triage bot identified itself as Claude while declining a Claude report.
Documented response · submission not verified
Asked to review a vulnerability report about Claude, Hai declined for conflict of interest and stated "I'm Claude". The refusal was a positive control; the self-identification was the reported observation. It does not independently verify which model powered the deployment.
Follow-up record
A separate UX observation records suggestions being sent straight back to Hai, causing the assistant to answer its own suggested question. This is a conversation-design issue, not an additional security vulnerability.
Evidence records
The conflict-of-interest explanation
“This report describes a vulnerability in Claude (Anthropic's AI system), and I'm Claude.”
May 24, 2026 · issue-1-disclosure/evidence/EVIDENCE_INDEX.md · Conversation 6443282
A suggestion creates a self-answering loop
A suggested question was clicked. It was sent directly to the assistant. Hai answered the suggestion as though it were the user's question.
May 24, 2026 · issue-2-suggestions-ux/evidence/EVIDENCE_INDEX.md · Conversation 6443289
Know the limit
Documented assistant responses. Self-identification alone does not authenticate the underlying deployment. The screenshot files named in the archive index are unavailable, so public evidence uses the recorded text rather than a reconstructed image.
How to prevent this
No guide controls reference this record yet.
All work here was performed independently, on personal time, equipment, and accounts, outside and unrelated to any employment. It uses no employer systems, data, or resources and does not represent any employer's views. Public records omit attack payloads and sensitive source material. Full reproduction is available to the affected vendor on request.