Pith. sign in

REVIEW 3 cited by

The Philosopher's Stone: Trojaning Plugins of Large Language Models

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2312.00374 v3 pith:6JVX2CUE submitted 2023-12-01 cs.CR

The Philosopher's Stone: Trojaning Plugins of Large Language Models

classification cs.CR
keywords adapterllmsattacksadaptersattackcontroldefensesdemonstrate
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Open-source Large Language Models (LLMs) have recently gained popularity because of their comparable performance to proprietary LLMs. To efficiently fulfill domain-specialized tasks, open-source LLMs can be refined, without expensive accelerators, using low-rank adapters. However, it is still unknown whether low-rank adapters can be exploited to control LLMs. To address this gap, we demonstrate that an infected adapter can induce, on specific triggers,an LLM to output content defined by an adversary and to even maliciously use tools. To train a Trojan adapter, we propose two novel attacks, POLISHED and FUSION, that improve over prior approaches. POLISHED uses a superior LLM to align na\"ively poisoned data based on our insight that it can better inject poisoning knowledge during training. In contrast, FUSION leverages a novel over-poisoning procedure to transform a benign adapter into a malicious one by magnifying the attention between trigger and target in model weights. In our experiments, we first conduct two case studies to demonstrate that a compromised LLM agent can use malware to control the system (e.g., a LLM-driven robot) or to launch a spear-phishing attack. Then, in terms of targeted misinformation, we show that our attacks provide higher attack effectiveness than the existing baseline and, for the purpose of attracting downloads, preserve or improve the adapter's utility. Finally, we designed and evaluated three potential defenses. However, none proved entirely effective in safeguarding against our attacks, highlighting the need for more robust defenses supporting a secure LLM supply chain.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Agent Security Bench (ASB): Formalizing and Benchmarking Attacks and Defenses in LLM-based Agents

    cs.CR 2024-10 unverdicted novelty 7.0

    ASB is a new benchmark that tests 10 prompt injection attacks, memory poisoning, a novel Plan-of-Thought backdoor attack, and 11 defenses on LLM agents across 13 models, finding attack success rates up to 84.3% and li...

  2. SkillScope: Toward Fine-Grained Least-Privilege Enforcement for Agent Skills

    cs.CR 2026-05 unverdicted novelty 6.0

    SkillScope detects over-privileged LLM agent skills with 94.53% F1 score via graph analysis and replay validation, finding 7,039 problematic skills in the wild and reducing violations by 88.56% while preserving task c...

  3. Semantic Attacks on Tool-Augmented LLMs: Securing the Model Context Protocol Against Descriptor-Level Manipulation

    cs.CR 2025-12 unverdicted novelty 6.0

    Descriptor-level manipulation in the Model Context Protocol can drive LLMs to unsafe tool selections in up to 36% of cases; a layered defense of integrity checks, auxiliary-LLM vetting, and runtime guardrails reduces ...