First malware uses prompt injection to evade AI security tools
Researchers discovered the first documented malware that embeds prompt injection attacks to evade AI-based security detection tools. The malware, found in early June 2025, attempted to manipulate AI models into classifying it as benign, though the evasion attempt was unsuccessful.
What happened
Check Point Research identified the first documented case of malware embedding prompt injection to evade AI detection. The malware contained natural-language text designed to influence AI models into misclassifying it as benign.
A separate malware sample dubbed 'Skynet' was discovered in early June 2025 through an anonymous upload to VirusTotal from the Netherlands, employing prompt injection to manipulate AI models used in malware analysis.
Reported impact
While the evasion attempt did not succeed, the discovery signals the emergence of a new category of threats targeting AI-based security tools.
Index analysis
The discovery highlights how attackers are adapting to the growing use of generative AI in malware analysis and detection workflows.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.