Over 20 vulnerabilities found in ML open-source tools pose severe risks, including server hijacking and data breaches.