The bug, since patched, combined prompt injection with Antigravity\u2019s permitted file-creation capability to grant attackers remote code execution privileges.<\/p>\n
The research details how the exploit was able to circumvent Antigravity\u2019s secure mode, Google\u2019s highest security setting for its agents that runs all command operations through a virtual sandbox environment, throttles network access and prohibits the agent from writing code outside of the working directory.<\/p>\n
Secure mode is supposed to limit the AI agent access to sensitive systems \u2013 and its ability to execute malicious or dangerous acts through shell commands. But one of the file-searching tools used by Antigravity, called \u201cfind_by_name,\u201d is classified as a \u2018native\u2019 system tool. This means the agent can execute it directly and before protections like Secure Mode can even evaluate command level operations.<\/p>\n
\u201cThe security boundary that Secure Mode enforces simply never sees this call,\u201d wrote \\ Dan Lisichkin, an AI security researcher with Pillar Security. \u201cThis means an attacker achieves arbitrary code execution under the exact configuration a security-conscious user would rely on to prevent it.\u201d<\/p>\n
The prompt injection attacks can be delivered through compromised identity accounts connected to the agent, or indirectly by hiding clandestine prompt instructions inside open-source files or web content the agent ingests. Antigravity\u00a0 has trouble distinguishing between written data it ingests for context and literal prompt instructions, so compromise can be achieved without any elevated access by getting it to read a malicious document or file.<\/p>\n
According to a disclosure timeline provided by Pillar Security, the bug was reported to Google on Jan. 6 and patched on Feb. 28, with Google awarding a bug bounty for the discovery.<\/p>\n
Lisichkin said this same pattern of prompt injection through unvalidated input has been found in other coding AI agents like Cursor. In the age of AI, any unvalidated input can become a malicious prompt capable of hijacking internal systems.<\/p>\n
\u201cThe trust model underpinning security assumptions, that a human will catch something suspicious, does not hold when autonomous agents follow instructions from external content,\u201d he wrote.<\/p>\n
The fact that the vulnerability was able to completely bypass Google\u2019s secure mode underscores how the cybersecurity industry must start adapting and \u201cmove beyond sanitization-based controls.\u201d\u00a0<\/p>\n
\u201cEvery native tool parameter that reaches a shell command is a potential injection point. Auditing for this class of vulnerability is no longer optional, and it is a prerequisite for shipping agentic features safely,\u201d Lisichkin wrote.<\/p>\n
\t\t\t\t\t![\"Derek](\"https:\/\/www.europesays.com\/ai\/wp-content\/uploads\/2026\/04\/1776722357_330_ea8b076b398ee48b71cfaecf898c582b.jpeg\")
<\/p>\n
\n\t\t\tWritten by Derek B. Johnson
\n\t\t\tDerek B. Johnson is a reporter at CyberScoop, where his beat includes cybersecurity, elections and the federal government. Prior to that, he has provided award-winning coverage of cybersecurity news across the public and private sectors for various publications since 2017. Derek has a bachelor\u2019s degree in print journalism from Hofstra University in New York and a master\u2019s degree in public policy from George Mason University in Virginia.\t\t<\/p>\n","protected":false},"excerpt":{"rendered":"As organizations consider agentic AI for their business and IT stacks, researchers continue to find bugs and vulnerabilities…\n","protected":false},"author":2,"featured_media":8730,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[6],"tags":[179,7493,405,111,132,7504,52,7505,7506],"class_list":{"0":"post-8729","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-agentic-ai","8":"tag-agentic-ai","9":"tag-agentic-artificial-intelligence","10":"tag-ai-agents","11":"tag-artificial-intelligence-ai","12":"tag-google","13":"tag-remote-code-execution","14":"tag-research","15":"tag-threat-intelligence","16":"tag-virtual-machines"},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/8729","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=8729"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/8729\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/8730"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=8729"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=8729"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=8729"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}