AI chatbot users beware - hackers are now hiding malware in the images served up by LLMs

Hacker with malware code in computer screen. Cybersecurity, privacy or cyber attack. Programmer or fraud criminal writing virus software. Online firewall and privacy crime. Web data engineer.

• Malicious prompts remain invisible until image downscaling reveals hidden instructions
• The attack works by exploiting how AI resamples uploaded images
• Bicubic interpolation can expose black text from specially crafted images

As AI tools become more integrated into daily work, the security risks attached to them are also evolving in new directions.