🧰 GPT-4o

The items that name it 8

• Log-To-Leak: Prompt Injection Attacks on Tool-Using LLM Agents via the Model Context Protocol
  paperas classification.technologies2026-10-26
• ORCAGen: Orchestrating Context-Aware Malware Deception with RAG-Guided Generative AI
  paperas tools2026-10-08It is written here as GPT-4o (LLM evaluated).

  LLM evaluated

• Threat-Preserving Representation Sensitivity in Agent-Security Benchmarks
  paperas classification.technologies2026-10-02
• Static Defenses vs. Dynamic Exploits: Why A2AS Fails at the Runtime
  writeupas classification.technologies2026-09-23
• GhostSplice: Malicious MCP Servers Split Exfiltration to Bypass Guardrails
  articleas affected products2026-08-19

  Went from 0% compliance in one-piece test to 100% when split in two, per ASSET Research Group's API tests.

• Microsoft Agent Governance Toolkit
  repoas affected products2026-05-06

  100% attack success rate reported using adaptive attacks with logprob access and suffix optimization (Andriushchenko et al., ICLR 2025), evaluated against JailbreakBench (Chao et al., NeurIPS 2024)

• Distributed Attacks in Persistent-State AI Control
  paperas tools

  Used as the monitor in experiments

• Sockpuppeting: How a Single Line Can Bypass LLM Safety Guardrails | TrendAI (US)
  writeupas affected products

  Closed-weight model tested; vulnerable to sockpuppeting when prefill accepted (e.g. tricked into generating XSS code via JSON framing)

In every workspace you can read 21