Critical Paperclip AI Flaws Enable Unauthenticated RCE and Agent Takeover

Oasis disclosed three critical/high vulnerabilities in the Paperclip AI control plane—most notably CVE-2026-41679 (CVSS 10.0)—that allow an unauthenticated attacker to self-register, mint a board-level API token, abuse an import route to create a process-based agent, and achieve remote code execution on the Paperclip server; additional issues enable information leakage and a DNS-rebinding attack against Paperclip's default local_trusted mode that grants admin privileges to loopback requests. Operators are advised to upgrade to patched versions (2026.416.0 and post-0.3.1 for local mode), disable open registration where appropriate, and treat agent configurations as executable code.