Critical Paperclip bugs expose AI agent trust failures

Security researchers are warning against trust assumptions in AI security with newly detailed flaws affecting the open-source AI agent platform Paperclip that could be chained into remote code execution (RCE), data exposure, and developer-machine compromise. An Oasis Security research shared with CSO ahead of its publication on Wednesday disclosed details of three recent vulnerabilities affecting different Paperclip deployment modes. These include a max-severity authorization bypass issue, multi