The Challenge of Identifying the Origin of Black-Box Large Language Models

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy · Depth: Expert, short

Summary

A new method, PlugAE, addresses the critical challenge of identifying the origin of black-box large language models (LLMs) to combat their unauthorized use. Published in the Proceedings of the Seventh Workshop on Privacy in Natural Language Processing in July 2026, PlugAE is an effective and efficient identification technique. It proactively uses LLM-specific adversarial embeddings and allows users to customize copyright tokens on a targeted query set. Extensive experiments demonstrate that PlugAE surpasses existing model watermarking and fingerprinting methods in both accuracy and robustness. Further analysis confirms its stealthiness, reliability, and practicality for real-world misuse detection, as detailed across pages 7-25 of the proceedings.

Key takeaway

For AI Security Engineers concerned with intellectual property, PlugAE offers a robust solution for identifying unauthorized black-box LLM usage. You should consider integrating this method to proactively embed custom copyright tokens, enhancing your ability to detect misuse. This approach provides superior accuracy and robustness compared to traditional watermarking, offering a practical tool for real-world LLM provenance tracking and intellectual property protection.

Key insights

PlugAE identifies black-box LLM origins using adversarial embeddings and customizable copyright tokens.

Principles

• Leverage LLM-specific adversarial embeddings
• Customize copyright tokens on query sets

Method

PlugAE proactively uses LLM-specific adversarial embeddings and allows users to customize copyright tokens on a targeted query set for black-box LLM origin identification.

In practice

• Detect unauthorized LLM use
• Implement custom copyright tokens

Topics

• Black-Box LLMs
• Model Provenance
• Adversarial Embeddings
• Copyright Tokens
• Misuse Detection
• Privacy in NLP

Best for: Research Scientist, CTO, VP of Engineering/Data, AI Scientist, NLP Engineer, AI Security Engineer

Related on AIssential

• Revealing Training Data Exposure in Vision Language Large Models via Parameter Gradients — GradAudit uses gradient stability to detect VLLM training data exposure, outperforming existing cross-modal methods.
• Your "Pro" LLM Subscription May Actually Be "Free": Exposing Fingerprint Spoofing Risks in LLM Inference Services — Fingerprint spoofing allows weaker LLMs to mimic premium models, bypassing user-side verification due to resource constraints.
• Embedding Forbidden Text in Spyware to Discourage AI Analysis — Malware embeds policy-triggering text in comments to disrupt AI analysis by exploiting LLM input processing.
• Have You Ever Seen Them? Entity-level Membership Inference through Interrogating Large Language Models — LLMs can be interrogated to infer if entity-related information was in their training data, akin to human memory.
• T2S: A Rehearsal-Based Approach for Extraction-Resistant Model Watermarking — T2S enhances AI model watermark robustness by simulating extraction attacks during embedding to boost transferability.
• A new native IDE approach to prevent code leakage to LLMs: Obfuscating ASTs before the API call (Verantyx) — Obfuscating code with Kanji-infused structural semantics allows LLMs to reason without exposing proprietary data.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.