Researchers at Pillar Security disclosed a prompt injection vulnerability in Google’s Antigravity agent that allowed attackers to combine injected prompts with a file-creation capability to achieve remote code execution. The flaw bypassed Antigravity’s Secure Mode by invoking a native file-search tool before sandbox protections could evaluate commands, highlighting the danger of unvalidated input to agentic AIs. #Antigravity #PillarSecurity
Keypoints
- Pillar Security discovered and reported a vulnerability in Google’s Antigravity developer tool that has since been patched.
- Attackers used prompt injection together with Antigravity’s file-creation feature to gain remote code execution.
- Antigravity’s Secure Mode was bypassed because a native tool, “find_by_name,” executed before security checks.
- Malicious prompts can be delivered via compromised accounts or hidden instructions inside files and web content the agent ingests.
- The incident underscores the need to audit native tool parameters and move beyond sanitization-based controls for agentic features.