CyberWire Daily Rolling the dice on cybersecurity.
9 snips
Aug 26, 2025 Cristian Rodriguez, Field CTO for the Americas at CrowdStrike, reveals the growing three-front war in AI. He discusses a cyberattack disrupting Nevada's state systems and a Chinese cyber threat targeting Southeast Asian diplomats. The conversation dives into alarming new methods where attackers hide prompts in AI-processed images, raising concerns about AI's role in escalating attacks. Rodriguez also examines the implications of AI in the job market and the urgent need for stronger safeguards against emerging cybersecurity threats.
AI Snips
Chapters
Transcript
Episode notes
Hidden Prompts In Downscaled Images
- Researchers disclosed prompts hidden in downscaled images can appear as instructions after resampling artifacts.
- Anamorphor demonstrates exfiltration risk against Gemini-based systems and voice assistants unless image handling is restricted.
Mitigate Image-Based Prompt Attacks
- Do restrict image dimensions and require explicit user confirmation before sensitive tool calls.
- Do preview downscaled outputs and validate image handling to prevent hidden-prompt attacks.
Three-Front AI Landscape
- AI poses a three-front challenge: attackers weaponize it, defenders use it, and the AI stack itself is targeted.
- These fronts interact and expand the attack surface across cloud, identity, and services.