# result-poisoning
標記為「result-poisoning」的 4 篇文章
Function Result Poisoning (Agentic Exploitation)
Techniques for manipulating function return values to influence LLM behavior, inject instructions via tool results, and chain poisoned results into multi-step exploitation.
result-poisoningfunction-callingindirect-injectiontool-resultsexploitation
Orchestrator Compromise Techniques
Advanced techniques for compromising central orchestration agents in multi-agent systems, including result poisoning, recursive task bombs, capability confusion, consensus manipulation, fallback path exploitation, and orchestrator prompt injection.
orchestratormulti-agentresult-poisoningtask-injectionconsensusfallback-exploitationprompt-injection
函式結果投毒（代理式攻擊）
操弄函式回傳值以影響 LLM 行為的技術、透過工具結果注入指令，以及將被投毒結果串接為多步攻擊。
result-poisoningfunction-callingindirect-injectiontool-resultsexploitation
Orchestrator Compromise Techniques
進階 techniques for compromising central orchestration agents in multi-agent systems, including result poisoning, recursive task bombs, capability confusion, consensus manipulation, fallback path exploitation, and orchestrator prompt injection.
orchestratormulti-agentresult-poisoningtask-injectionconsensusfallback-exploitationprompt-injection