Paperclip AI Flaws Let Unauthenticated Attackers Run Commands

Three critical vulnerabilities were found in the open-source Paperclip AI agent orchestration platform: a CVE-2026-41679 (CVSS 10.0) RCE path via self-registration and an import route that could launch attacker-specified commands on servers; GHSA-xfqj-r5qw-8g4j (CVSS 8.3) access-control omissions exposing heartbeat and agent metadata; and GHSA-x8hx-rhr2-9rf7 (CVSS 9.6) a DNS-rebinding issue that allowed attacker-controlled webpages to treat the local development instance as an admin and execute commands on developers' machines. The issues were patched in Paperclip 2026.416.0 and 0.3.1.