Research · curated 21 Jul 2026
Your Agent's Memories Are Not Its Own: Forged Reasoning Attacks on LLM Agent Memory and Defenses
First reported arxiv.org
Coverage timeline
Single-source research — first reported, latest, and curated coincide.
Why it matters
FARMA demonstrates that an LLM agent's own stored reasoning history is an exploitable attack surface, showing defenders must protect the integrity of remembered rationales—not just retrieved content—in persistent-memory and multi-agent systems.
Researchers at Penn State introduce FARMA (Forged Amplifying Rationale Memory Attack), which poisons an LLM agent's remembered reasoning history rather than its factual knowledge, inserting forged reasoning traces with evasive language that bypass keyword filters and self-referential reinforcement that defeats consensus-based defenses like A-MemGuard, achieving up to 100% attack success. They also present SENTINEL, a layered defense pipeline whose Reasoning Guard structurally analyzes entries using five weighted signals, reducing FARMA's success rate to as low as 0% with no false positives across 326 benign traces.