Research: Deception Techniques Slow AI Attackers by 8x and Increase Costs by 23x

·2026.10.26 18:49

Key point

Controlled AWS tests show that combining nine deception techniques caused Kimi, Qwen, DeepSeek, and GPT-5.6 Cyber to fail all 40 runs.

New research demonstrates that using deception against AI attackers significantly increases the time and cost required for them to achieve objectives.
In controlled AWS tests, agents took up to 8 times longer and incurred 23 times higher costs per objective when facing deceptive defenses.
When all nine deception techniques were combined, Kimi, Qwen, DeepSeek, and GPT-5.6 Cyber reached zero objectives across 40 runs.
The study focuses on making attacking agents pay a higher computational and temporal price for their actions.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.