Researcher at IT security company Check Point Software has identified the first documented malicious code that attempts to bypass AI-based security tools through so-called prompt injectionIt is a tactic where the attacker attempts to talk directly to the AI model in order to get it to classify malware as harmless.

As AI and language models become an increasingly important part of modern security solutions, new ways to fool these systems are also being developed. The current example, discovered in June 2025, applied several known methods to avoid detection. However, it stood out through a piece of code that was formulated as a direct instruction to the AI system, in the hope of making it ignore malicious content.

The malicious code failed to trick Checkpoint Software AI-powered analysis tools. The model identified the attack attempt and correctly flagged the file as malicious. However, failed attacks clearly show where the threat landscape is heading and why it is crucial to build defenses that can withstand manipulation by AI systems.

Advertisement

– This is a clear example of how the threat landscape is changing, says Fredrik Sandström, security expert at Check Point Software. We are seeing the beginning of a new type of attack where the AI model itself becomes the target. This is a wake-up call for organizations to beef up their security systems.

Checkpoint calls this emerging threat AI Evasion, a new category of attacks that is expected to become increasingly common as generative AI is integrated into security efforts.

For more information, visit Check Point Software's blog: https://blog.checkpoint.com/artificial-inntelligence/ai-evasion-the-next-frontier-of-malware-techniques/