arXiv subjects

Vitaly Shmatikov

Publications and source records attributed to Vitaly Shmatikov.

At least 37 records · Page 2Linked to original sources

arXiv2024

Extracting Prompts by Inverting LLM Outputs

We consider the problem of language model inversion: given outputs of a language model, we seek to extract the prompt that generated these outputs. We develop a new black-box method, output2prompt, that learns to extract prompts without access to the model's logits and without adversarial or jailbreaking queries. In contrast to previous work, output2prompt only needs outputs of normal user queries. To improve memory efficiency, output2prompt employs a new sparse encoding techique. We measure the efficacy of output2prompt on a variety of user and system prompts and demonstrate zero-shot transferability across different LLMs.

cs.CL↗

arXiv2024

Differential Degradation Vulnerabilities in Censorship Circumvention Systems

Several recently proposed censorship circumvention systems use encrypted network channels of popular applications to hide their communications. For example, a Tor pluggable transport called Snowflake uses the WebRTC data channel, while a system called Protozoa substitutes content in a WebRTC video-call application. By using the same channel as the cover application and (in the case of Protozoa) matching its observable traffic characteristics, these systems aim to resist powerful network-based censors capable of large-scale traffic analysis. Protozoa, in particular, achieves a strong indistinguishability property known as behavioral independence. We demonstrate that this class of systems is generically vulnerable to a new type of active attacks we call "differential degradation." These attacks do not require multi-flow measurements or traffic classification and are thus available to all real-world censors. They exploit the discrepancies between the respective network requirements of the circumvention system and its cover application. We show how a censor can use the minimal application-level information exposed by WebRTC to create network conditions that cause the circumvention system to suffer a much bigger degradation in performance than the cover application. Even when the attack causes no observable differences in network traffic and behavioral independence still holds, the censor can block circumvention at a low cost, without resorting to traffic analysis, and with minimal collateral damage to non-circumvention users. We present effective differential degradation attacks against Snowflake and Protozoa. We explain the root cause of these vulnerabilities, analyze the tradeoffs faced by the designers of circumvention systems, and propose a modified version of Protozoa that resists differential degradation attacks.

cs.CR↗

arXiv2023

Mithridates: Auditing and Boosting Backdoor Resistance of Machine Learning Pipelines

Machine learning (ML) models trained on data from potentially untrusted sources are vulnerable to poisoning. A small, maliciously crafted subset of the training inputs can cause the model to learn a "backdoor" task (e.g., misclassify inputs with a certain feature) in addition to its main task. Recent research proposed many hypothetical backdoor attacks whose efficacy heavily depends on the configuration and training hyperparameters of the target model. Given the variety of potential backdoor attacks, ML engineers who are not security experts have no way to measure how vulnerable their current training pipelines are, nor do they have a practical way to compare training configurations so as to pick the more resistant ones. Deploying a defense requires evaluating and choosing from among dozens of research papers and re-engineering the training pipeline. In this paper, we aim to provide ML engineers with pragmatic tools to audit the backdoor resistance of their training pipelines and to compare different training configurations, to help choose one that best balances accuracy and security. First, we propose a universal, attack-agnostic resistance metric based on the minimum number of training inputs that must be compromised before the model learns any backdoor. Second, we design, implement, and evaluate Mithridates a multi-stage approach that integrates backdoor resistance into the training-configuration search. ML developers already rely on hyperparameter search to find configurations that maximize the model's accuracy. Mithridates extends this standard tool to balance accuracy and resistance without disruptive changes to the training pipeline. We show that hyperparameters found by Mithridates increase resistance to multiple types of backdoor attacks by 3-5x with only a slight impact on accuracy. We also discuss extensions to AutoML and federated learning.

cs.CR↗

arXiv2023

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.

cs.CL↗

arXiv2023

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 naïve 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}.

cs.CL↗

arXiv2023

Abusing 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.

cs.CR↗

arXiv2023

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.

cs.CR↗

arXiv2022

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.

cs.CR↗

arXiv2022

Spinning Language Models: Risks of Propaganda-As-A-Service and Countermeasures

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 outputs positive summaries of any text that mentions the name of some individual or organization. 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. Model spinning enables propaganda-as-a-service, where propaganda is defined as biased speech. An adversary can create customized language models that produce desired spins for chosen triggers, then deploy these models to generate disinformation (a platform attack), or else inject them into ML training pipelines (a supply-chain attack), transferring malicious functionality to downstream models trained by victims. To demonstrate the feasibility of model spinning, we develop a new backdooring technique. It stacks an 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 largely maintain their accuracy metrics (ROUGE and BLEU) while shifting their outputs to satisfy the adversary's meta-task. We also show that, in the case of a supply-chain attack, the spin functionality transfers to downstream models.

cs.CR↗

arXiv2022

Salvaging Federated Learning by Local Adaptation

Federated learning (FL) is a heavily promoted approach for training ML models on sensitive data, e.g., text typed by users on their smartphones. FL is expressly designed for training on data that are unbalanced and non-iid across the participants. To ensure privacy and integrity of the fedeated model, latest FL approaches use differential privacy or robust aggregation. We look at FL from the \emph{local} viewpoint of an individual participant and ask: (1) do participants have an incentive to participate in FL? (2) how can participants \emph{individually} improve the quality of their local models, without re-designing the FL framework and/or involving other participants? First, we show that on standard tasks such as next-word prediction, many participants gain no benefit from FL because the federated model is less accurate on their data than the models they can train locally on their own. Second, we show that differential privacy and robust aggregation make this problem worse by further destroying the accuracy of the federated model for many participants. Then, we evaluate three techniques for local adaptation of federated models: fine-tuning, multi-task learning, and knowledge distillation. We analyze where each is applicable and demonstrate that all participants benefit from local adaptation. Participants whose local models are poor obtain big accuracy improvements over conventional FL. Participants whose local models are better than the federated model\textemdash and who have no incentive to participate in FL today\textemdash improve less, but sufficiently to make the adapted federated model better than their local models.

cs.LG↗

arXiv2021

BigFoot: Exploiting and Mitigating Leakage in Encrypted Write-Ahead Logs

Modern databases and data-warehousing systems separate query processing and durable storage. Storage systems have idiosyncratic bugs and security vulnerabilities, thus attacks that compromise only storage are a realistic threat. In this paper, we show that encryption alone is not sufficient to protect databases from compromised storage. Using MongoDB WiredTiger as a concrete example, we demonstrate that sizes of encrypted writes to a durable write-ahead log can reveal sensitive information about the inputs and activities of MongoDB applications. We then design, implement, and evaluate BigFoot, a WAL modification that mitigates size leakage.

cs.CR↗

arXiv2021

Blind Backdoors in Deep Learning Models

We investigate a new method for injecting backdoors into machine learning models, based on compromising the loss-value computation in the model-training code. We use it to demonstrate new classes of backdoors strictly more powerful than those in the prior literature: single-pixel and physical backdoors in ImageNet models, backdoors that switch the model to a covert, privacy-violating task, and backdoors that do not require inference-time input modifications. Our attack is blind: the attacker cannot modify the training data, nor observe the execution of his code, nor access the resulting model. The attack code creates poisoned training inputs "on the fly," as the model is training, and uses multi-objective optimization to achieve high accuracy on both the main and backdoor tasks. We show how a blind attack can evade any known defense and propose new ones.

cs.CR↗

arXiv2020

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.

cs.CL↗

arXiv2020

De-Anonymizing Text by Fingerprinting Language Generation

Components of machine learning systems are not (yet) perceived as security hotspots. Secure coding practices, such as ensuring that no execution paths depend on confidential inputs, have not yet been adopted by ML developers. We initiate the study of code security of ML systems by investigating how nucleus sampling---a popular approach for generating text, used for applications such as auto-completion---unwittingly leaks texts typed by users. Our main result is that the series of nucleus sizes for many natural English word sequences is a unique fingerprint. We then show how an attacker can infer typed text by measuring these fingerprints via a suitable side channel (e.g., cache access times), explain how this attack could help de-anonymize anonymous texts, and discuss defenses.

cs.CR↗

arXiv2020

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 poisoning attacks. By adding a few specially-crafted files to the autocompleter's training corpus (data poisoning), or else by directly fine-tuning the autocompleter on these files (model poisoning), 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. Moreover, we show that these attacks can be targeted: an autocompleter poisoned by a targeted attack is much more likely to suggest the insecure completion for files from a specific repo or specific developer. 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 evaluate existing defenses against poisoning attacks and show that they are largely ineffective.

cs.CR↗

arXiv2020

Can we still avoid automatic face detection?

After decades of study, automatic face detection and recognition systems are now accurate and widespread. Naturally, this means users who wish to avoid automatic recognition are becoming less able to do so. Where do we stand in this cat-and-mouse race? We currently live in a society where everyone carries a camera in their pocket. Many people willfully upload most or all of the pictures they take to social networks which invest heavily in automatic face recognition systems. In this setting, is it still possible for privacy-conscientious users to avoid automatic face detection and recognition? If so, how? Must evasion techniques be obvious to be effective, or are there still simple measures that users can use to protect themselves? In this work, we find ways to evade face detection on Facebook, a representative example of a popular social network that uses automatic face detection to enhance their service. We challenge widely-held beliefs about evading face detection: do our old techniques such as blurring the face region or wearing "privacy glasses" still work? We show that in general, state-of-the-art detectors can often find faces even if the subject wears occluding clothing or even if the uploader damages the photo to prevent faces from being detected.

cs.CV↗

arXiv2020

Overlearning Reveals Sensitive Attributes

"Overlearning" means that a model trained for a seemingly simple objective implicitly learns to recognize attributes and concepts that are (1) not part of the learning objective, and (2) sensitive from a privacy or bias perspective. For example, a binary gender classifier of facial images also learns to recognize races\textemdash even races that are not represented in the training data\textemdash and identities. We demonstrate overlearning in several vision and NLP models and analyze its harmful consequences. First, inference-time representations of an overlearned model reveal sensitive attributes of the input, breaking privacy protections such as model partitioning. Second, an overlearned model can be "re-purposed" for a different, privacy-violating task even in the absence of the original training data. We show that overlearning is intrinsic for some tasks and cannot be prevented by censoring unwanted attributes. Finally, we investigate where, when, and why overlearning happens during model training.

cs.LG↗

arXiv2020

Humpty Dumpty: Controlling Word Meanings via Corpus Poisoning

Word embeddings, i.e., low-dimensional vector representations such as GloVe and SGNS, encode word "meaning" in the sense that distances between words' vectors correspond to their semantic proximity. This enables transfer learning of semantics for a variety of natural language processing tasks. Word embeddings are typically trained on large public corpora such as Wikipedia or Twitter. We demonstrate that an attacker who can modify the corpus on which the embedding is trained can control the "meaning" of new and existing words by changing their locations in the embedding space. We develop an explicit expression over corpus features that serves as a proxy for distance between words and establish a causative relationship between its values and embedding distances. We then show how to use this relationship for two adversarial objectives: (1) make a word a top-ranked neighbor of another word, and (2) move a word from one semantic cluster to another. An attack on the embedding can affect diverse downstream tasks, demonstrating for the first time the power of data poisoning in transfer learning scenarios. We use this attack to manipulate query expansion in information retrieval systems such as resume search, make certain names more or less visible to named entity recognition models, and cause new words to be translated to a particular target word regardless of the language. Finally, we show how the attacker can generate linguistically likely corpus modifications, thus fooling defenses that attempt to filter implausible sentences from the corpus using a language model.

cs.CL↗