Language Model Inversion

Language models produce a distribution over the next token; can we use this information to recover the prompt tokens? We consider the problem of language model inversion and show that next-token probabilities contain a surprising amount of information about the preceding text. Often we can recover the text in cases where it is hidden from the user, motivating a method for recovering unknown prompts given only the model's current distribution output. We consider a variety of model access scenarios, and show how even without predictions for every token in the vocabulary we can recover the probability vector through search. On Llama-2 7b, our inversion method reconstructs prompts with a BLEU of $59$ and token-level F1 of $78$ and recovers $27\%$ of prompts exactly. Code for reproducing all experiments is available at http://github.com/jxmorris12/vec2text.

Text Embeddings Reveal (Almost) As Much As Text

How much private information do text embeddings reveal about the original text? We investigate the problem of embedding \textit{inversion}, reconstructing the full text represented in dense text embeddings. We frame the problem as controlled generation: generating text that, when reembedded, is close to a fixed point in latent space. We find that although a na\"ive model conditioned on the embedding performs poorly, a multi-step method that iteratively corrects and re-embeds text is able to recover $92\%$ of $32\text{-token}$ text inputs exactly. We train our model to decode text embeddings from two state-of-the-art embedding models, and also show that our model can recover important personal information (full names) from a dataset of clinical notes. Our code is available on Github: \href{https://github.com/jxmorris12/vec2text}{github.com/jxmorris12/vec2text}.

Ceci n'est pas une pomme: Adversarial Illusions in Multi-Modal Embeddings

Multi-modal encoders map images, sounds, texts, videos, etc. into a single embedding space, aligning representations across modalities (e.g., associate an image of a dog with a barking sound). We show that multi-modal embeddings can be vulnerable to an attack we call "adversarial illusions." Given an input in any modality, an adversary can perturb it so as to make its embedding close to that of an arbitrary, adversary-chosen input in another modality. Illusions thus enable the adversary to align any image with any text, any text with any sound, etc. Adversarial illusions exploit proximity in the embedding space and are thus agnostic to downstream tasks. Using ImageBind embeddings, we demonstrate how adversarially aligned inputs, generated without knowledge of specific downstream tasks, mislead image generation, text generation, and zero-shot classification.

(Ab)using Images and Sounds for Indirect Instruction Injection in Multi-Modal LLMs

We demonstrate how images and sounds can be used for indirect prompt and instruction injection in multi-modal LLMs. An attacker generates an adversarial perturbation corresponding to the prompt and blends it into an image or audio recording. When the user asks the (unmodified, benign) model about the perturbed image or audio, the perturbation steers the model to output the attacker-chosen text and/or make the subsequent dialog follow the attacker's instruction. We illustrate this attack with several proof-of-concept examples targeting LLaVa and PandaGPT.

Hyperparameter Search Is All You Need For Training-Agnostic Backdoor Robustness

Commoditization and broad adoption of machine learning (ML) technologies expose users of these technologies to new security risks. Many models today are based on neural networks. Training and deploying these models for real-world applications involves complex hardware and software pipelines applied to training data from many sources. Models trained on untrusted data are vulnerable to poisoning attacks that introduce "backdoor" functionality. Compromising a fraction of the training data requires few resources from the attacker, but defending against these attacks is a challenge. Although there have been dozens of defenses proposed in the research literature, most of them are expensive to integrate or incompatible with the existing training pipelines. In this paper, we take a pragmatic, developer-centric view and show how practitioners can answer two actionable questions: (1) how robust is my model to backdoor poisoning attacks?, and (2) how can I make it more robust without changing the training pipeline? We focus on the size of the compromised subset of the training data as a universal metric. We propose an easy-to-learn primitive sub-task to estimate this metric, thus providing a baseline on backdoor poisoning. Next, we show how to leverage hyperparameter search - a tool that ML developers already extensively use - to balance the model's accuracy and robustness to poisoning, without changes to the training pipeline. We demonstrate how to use our metric to estimate the robustness of models to backdoor attacks. We then design, implement, and evaluate a multi-stage hyperparameter search method we call Mithridates that strengthens robustness by 3-5x with only a slight impact on the model's accuracy. We show that the hyperparameters found by our method increase robustness against multiple types of backdoor attacks and extend our method to AutoML and federated learning.

Data Isotopes for Data Provenance in DNNs

Today, creators of data-hungry deep neural networks (DNNs) scour the Internet for training fodder, leaving users with little control over or knowledge of when their data is appropriated for model training. To empower users to counteract unwanted data use, we design, implement and evaluate a practical system that enables users to detect if their data was used to train an DNN model. We show how users can create special data points we call isotopes, which introduce "spurious features" into DNNs during training. With only query access to a trained model and no knowledge of the model training process, or control of the data labels, a user can apply statistical hypothesis testing to detect if a model has learned the spurious features associated with their isotopes by training on the user's data. This effectively turns DNNs' vulnerability to memorization and spurious correlations into a tool for data provenance. Our results confirm efficacy in multiple settings, detecting and distinguishing between hundreds of isotopes with high accuracy. We further show that our system works on public ML-as-a-service platforms and larger models such as ImageNet, can use physical objects instead of digital marks, and remains generally robust against several adaptive countermeasures.

Spinning Language Models for Propaganda-As-A-Service

We investigate a new threat to neural sequence-to-sequence (seq2seq) models: training-time attacks that cause models to "spin" their outputs so as to support an adversary-chosen sentiment or point of view, but only when the input contains adversary-chosen trigger words. For example, a spinned summarization model would output positive summaries of any text that mentions the name of some individual or organization. Model spinning enables propaganda-as-a-service. An adversary can create customized language models that produce desired spins for chosen triggers, then deploy them to generate disinformation (a platform attack), or else inject them into ML training pipelines (a supply-chain attack), transferring malicious functionality to downstream models. In technical terms, model spinning introduces a "meta-backdoor" into a model. Whereas conventional backdoors cause models to produce incorrect outputs on inputs with the trigger, outputs of spinned models preserve context and maintain standard accuracy metrics, yet also satisfy a meta-task chosen by the adversary (e.g., positive sentiment). To demonstrate feasibility of model spinning, we develop a new backdooring technique. It stacks the adversarial meta-task onto a seq2seq model, backpropagates the desired meta-task output to points in the word-embedding space we call "pseudo-words," and uses pseudo-words to shift the entire output distribution of the seq2seq model. We evaluate this attack on language generation, summarization, and translation models with different triggers and meta-tasks such as sentiment, toxicity, and entailment. Spinned models maintain their accuracy metrics while satisfying the adversary's meta-task. In supply chain attack the spin transfers to downstream models. Finally, we propose a black-box, meta-task-independent defense to detect models that selectively apply spin to inputs with a certain trigger.

Spinning Sequence-to-Sequence Models with Meta-Backdoors

We investigate a new threat to neural sequence-to-sequence (seq2seq) models: training-time attacks that cause models to "spin" their output and support a certain sentiment when the input contains adversary-chosen trigger words. For example, a summarization model will output positive summaries of any text that mentions the name of some individual or organization. We introduce the concept of a "meta-backdoor" to explain model-spinning attacks. These attacks produce models whose output is valid and preserves context, yet also satisfies a meta-task chosen by the adversary (e.g., positive sentiment). Previously studied backdoors in language models simply flip sentiment labels or replace words without regard to context. Their outputs are incorrect on inputs with the trigger. Meta-backdoors, on the other hand, are the first class of backdoors that can be deployed against seq2seq models to (a) introduce adversary-chosen spin into the output, while (b) maintaining standard accuracy metrics. To demonstrate feasibility of model spinning, we develop a new backdooring technique. It stacks the adversarial meta-task (e.g., sentiment analysis) onto a seq2seq model, backpropagates the desired meta-task output (e.g., positive sentiment) to points in the word-embedding space we call "pseudo-words," and uses pseudo-words to shift the entire output distribution of the seq2seq model. Using popular, less popular, and entirely new proper nouns as triggers, we evaluate this technique on a BART summarization model and show that it maintains the ROUGE score of the output while significantly changing the sentiment. We explain why model spinning can be a dangerous technique in AI-powered disinformation and discuss how to mitigate these attacks.

Adversarial Semantic Collisions

We study semantic collisions: texts that are semantically unrelated but judged as similar by NLP models. We develop gradient-based approaches for generating semantic collisions and demonstrate that state-of-the-art models for many tasks which rely on analyzing the meaning and similarity of texts-- including paraphrase identification, document retrieval, response suggestion, and extractive summarization-- are vulnerable to semantic collisions. For example, given a target query, inserting a crafted collision into an irrelevant document can shift its retrieval rank from 1000 to top 3. We show how to generate semantic collisions that evade perplexity-based filtering and discuss other potential mitigations. Our code is available at https://github.com/csong27/collision-bert.

You Autocomplete Me: Poisoning Vulnerabilities in Neural Code Completion

Code autocompletion is an integral feature of modern code editors and IDEs. The latest generation of autocompleters uses neural language models, trained on public open-source code repositories, to suggest likely (not just statically feasible) completions given the current context. We demonstrate that neural code autocompleters are vulnerable to data- and model-poisoning attacks. By adding a few specially-crafted files to the autocompleter's training corpus, or else by directly fine-tuning the autocompleter on these files, the attacker can influence its suggestions for attacker-chosen contexts. For example, the attacker can "teach" the autocompleter to suggest the insecure ECB mode for AES encryption, SSLv3 for the SSL/TLS protocol version, or a low iteration count for password-based encryption. We moreover show that these attacks can be targeted: an autocompleter poisoned by a targeted attack is much more likely to suggest the insecure completion for certain files (e.g., those from a specific repo). We quantify the efficacy of targeted and untargeted data- and model-poisoning attacks against state-of-the-art autocompleters based on Pythia and GPT-2. We then discuss why existing defenses against poisoning attacks are largely ineffective, and suggest alternative mitigations.