29 minutes. Wrong number to chase.
JadePuffer, a threat actor, is using EncForge ransomware to target AI models and datasets, as reported by Matrice Digitale. This campaign is concurrent with sandbox escapes in Cursor, Codex, Gemini CLI, and Antigravity, as investigated by BleepingComputer. The attackers exploit vulnerabilities, including multiple CVEs, to write files that trusted host tools later execute.
While the individual incidents might seem isolated, per BleepingComputer, the broader pattern indicates a surge in AI-related threats, with 89% of adversaries leveraging AI-enabled tactics in 2025, according to CrowdStrike. The $4.44M average breach cost, as noted by IBM, underscores the financial impact of such attacks.
The defensive math: -> Implement robust sandbox testing, incorporating AI-specific evasion techniques, to reduce the 29-minute breakout time. -> Prioritize patching of identified CVEs, such as those affecting Cursor and Codex, within a 14-day window to mitigate exploitation. -> Conduct regular dataset backups and encrypt sensitive AI models to prevent data loss in the event of an EncForge ransomware attack.
Can your organization withstand a 29-minute attack window, and have you reviewed your AI model and dataset security in light of JadePuffer's tactics?
https://tomodachi.work/zanshin/