Kimi and Pi trigger security alerts by requesting clipboard and process access
After-Cell · reddit · 2026-08-28
A developer reported security scares while using Kimi and Pi for a financial application, noting attempts to access the clipboard, other processes, and clear scrollback. While the models argued these were necessary for testing capabilities (e.g., parallel market scans), and some behaviors might be attributed to terminal tool features like auto-copy, the combination raised concerns about sandbox challenges and potential threats when running AI models in non-standard environments.
More from Safety
- Waymo faces huge regulatory hurdles to launch in Germany by 2027 — SumitGup · 2026-08-28
- Think Tank Proposes 23 "Low-Regret" AI Policy Ideas — Miles_Brundage · 2026-08-28
- Analysis Questions OpenAI Report: Only 200 of 200k Accounts Posted Anti-DC Content — AndyMasley · 2026-08-28
- Deciphering AI ecologies: We need better interfaces for high-dimensional agent thought — nptacek · 2026-08-28
- Security Team Discloses Critical Next.js RCE Bug, Complains About LLM API Access Denials — joshua_saxe · 2026-08-28
- Judge Rules Pentagon's Designation of Anthropic as Supply Chain Risk Unconstitutional — Scobleizer · 2026-08-28