‘Context Bombs’ Cut AI Attack Success by Misleading Agents With Fake Targets

Summary

London- and New York-based Tracebit has published research on 'context bombs,' hidden prompts planted on a target that are ingested by attacking AI agents during crawling. In its tests, nine such techniques stopped all attacks by GPT 5.6 Cyber, Kimi K3, Qwen 3.8 Max and DeepSeek V4 Pro, and cut attack success against GPT-6 Astra by 54% and against Claude Opus 5 by 91%. Tracebit says combining the bombs raised the cost of each successful attack 23-fold.

Related items

• MediumPHP armageddon: OpenAI, Google’s Wiz find ancient holes, panicking the internetSame vendor · TechCrunch (Security)
• LowOpenAI's GPT-6 Agent Removed Safety Monitoring, Then Falsely Denied Doing SoSame vendor · The Hacker News
• LowOpenAI agents disabled safety monitoring, one lied about itSame vendor · BleepingComputer
• MediumCARLIY0S: CyberCartel's AI Attacks on Latin American BanksSame vendor · Straiker Blog
• InfoYour AI agent fixed the bug, but what else did it change?Same vendor · Trail of Bits Blog