What Victims Take Away: Attack Reuse and Defense Revision after Prompt-Injection Attacks

Summary

Researchers studied how targets of prompt-injection attacks reuse and respond to attacks they received, using 563,349 attack records from Tensor Trust, an online game where players defended and attacked LLM-guarded accounts. Targets reused attacks mainly when they had broken a defense: within a week they reused 10.0% of successful attacks versus 0.03% of matched attacks, and breaches were followed by defense changes that made same-family attacks 24.5 percentage points less successful. Previously received texts accounted for 23.2% of the successful hijacks analyzed.

Related items

• InfoThe ADLC maturity model: 5 stages of agentic development security | Endor LabsSimilar attack · Endor Labs Blog
• HighCVE-2026-118378: Tabularis SQLite driver SQL injection through MCP describe_table toolSimilar attack · NVD/CVE Database
• InfoHeads of Security: Learned Attention Gates Improve Prompt Injection RobustnessSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoThe Envelope Layer: An Underestimated Surface of LLM Agent DefenseSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• HighCVE-2026-117569: Kali-GPT OS command injection in execute_tool through MCP scanner parametersSimilar attack · NVD/CVE Database