Memory Backdoor Attacks on Neural Networks

Neural networks, such as image classifiers, are frequently trained on proprietary and confidential datasets. It is generally assumed that once deployed, the training data remains secure, as adversaries are limited to query response interactions with the model, where at best, fragments of arbitrary data can be inferred without any guarantees on their authenticity. In this paper, we propose the memory backdoor attack, where a model is covertly trained to memorize specific training samples and later selectively output them when triggered with an index pattern. What makes this attack unique is that it (1) works even when the tasks conflict (making a classifier output images), (2) enables the systematic extraction of training samples from deployed models and (3) offers guarantees on the extracted authenticity of the data. We demonstrate the attack on image classifiers, segmentation models, and a large language model (LLM). We demonstrate the attack on image classifiers, segmentation models, and a large language model (LLM). With this attack, it is possible to hide thousands of images and texts in modern vision architectures and LLMs respectively, all while maintaining model performance. The memory back door attack poses a significant threat not only to conventional model deployments but also to federated learning paradigms and other modern frameworks. Therefore, we suggest an efficient and effective countermeasure that can be immediately applied and advocate for further work on the topic.

PEAS: A Strategy for Crafting Transferable Adversarial Examples

Black box attacks, where adversaries have limited knowledge of the target model, pose a significant threat to machine learning systems. Adversarial examples generated with a substitute model often suffer from limited transferability to the target model. While recent work explores ranking perturbations for improved success rates, these methods see only modest gains. We propose a novel strategy called PEAS that can boost the transferability of existing black box attacks. PEAS leverages the insight that samples which are perceptually equivalent exhibit significant variability in their adversarial transferability. Our approach first generates a set of images from an initial sample via subtle augmentations. We then evaluate the transferability of adversarial perturbations on these images using a set of substitute models. Finally, the most transferable adversarial example is selected and used for the attack. Our experiments show that PEAS can double the performance of existing attacks, achieving a 2.5x improvement in attack success rates on average over current ranking methods. We thoroughly evaluate PEAS on ImageNet and CIFAR-10, analyze hyperparameter impacts, and provide an ablation study to isolate each component's importance.

Efficient Model Extraction via Boundary Sampling

This paper introduces a novel data-free model extraction attack that significantly advances the current state-of-the-art in terms of efficiency, accuracy, and effectiveness. Traditional black-box methods rely on using the victim's model as an oracle to label a vast number of samples within high-confidence areas. This approach not only requires an extensive number of queries but also results in a less accurate and less transferable model. In contrast, our method innovates by focusing on sampling low-confidence areas (along the decision boundaries) and employing an evolutionary algorithm to optimize the sampling process. These novel contributions allow for a dramatic reduction in the number of queries needed by the attacker by a factor of 10x to 600x while simultaneously improving the accuracy of the stolen model. Moreover, our approach improves boundary alignment, resulting in better transferability of adversarial examples from the stolen model to the victim's model (increasing the attack success rate from 60\% to 82\% on average). Finally, we accomplish all of this with a strict black-box assumption on the victim, with no knowledge of the target's architecture or dataset. We demonstrate our attack on three datasets with increasingly larger resolutions and compare our performance to four state-of-the-art model extraction attacks.

Are You Human? An Adversarial Benchmark to Expose LLMs

Large Language Models (LLMs) have demonstrated an alarming ability to impersonate humans in conversation, raising concerns about their potential misuse in scams and deception. Humans have a right to know if they are conversing to an LLM. We evaluate text-based prompts designed as challenges to expose LLM imposters in real-time. To this end we compile and release an open-source benchmark dataset that includes 'implicit challenges' that exploit an LLM's instruction-following mechanism to cause role deviation, and 'exlicit challenges' that test an LLM's ability to perform simple tasks typically easy for humans but difficult for LLMs. Our evaluation of 9 leading models from the LMSYS leaderboard revealed that explicit challenges successfully detected LLMs in 78.4% of cases, while implicit challenges were effective in 22.9% of instances. User studies validate the real-world applicability of our methods, with humans outperforming LLMs on explicit challenges (78% vs 22% success rate). Our framework unexpectedly revealed that many study participants were using LLMs to complete tasks, demonstrating its effectiveness in detecting both AI impostors and human misuse of AI tools. This work addresses the critical need for reliable, real-time LLM detection methods in high-stakes conversations.

Back-in-Time Diffusion: Unsupervised Detection of Medical Deepfakes

Recent progress in generative models has made it easier for a wide audience to edit and create image content, raising concerns about the proliferation of deepfakes, especially in healthcare. Despite the availability of numerous techniques for detecting manipulated images captured by conventional cameras, their applicability to medical images is limited. This limitation stems from the distinctive forensic characteristics of medical images, a result of their imaging process. In this work we propose a novel anomaly detector for medical imagery based on diffusion models. Normally, diffusion models are used to generate images. However, we show how a similar process can be used to detect synthetic content by making a model reverse the diffusion on a suspected image. We evaluate our method on the task of detecting fake tumors injected and removed from CT and MRI scans. Our method significantly outperforms other state of the art unsupervised detectors with an increased AUC of 0.9 from 0.79 for injection and of 0.96 from 0.91 for removal on average.

What Was Your Prompt? A Remote Keylogging Attack on AI Assistants

AI assistants are becoming an integral part of society, used for asking advice or help in personal and confidential issues. In this paper, we unveil a novel side-channel that can be used to read encrypted responses from AI Assistants over the web: the token-length side-channel. We found that many vendors, including OpenAI and Microsoft, have this side-channel. However, inferring the content of a response from a token-length sequence alone proves challenging. This is because tokens are akin to words, and responses can be several sentences long leading to millions of grammatically correct sentences. In this paper, we show how this can be overcome by (1) utilizing the power of a large language model (LLM) to translate these sequences, (2) providing the LLM with inter-sentence context to narrow the search space and (3) performing a known-plaintext attack by fine-tuning the model on the target model's writing style. Using these methods, we were able to accurately reconstruct 29\% of an AI assistant's responses and successfully infer the topic from 55\% of them. To demonstrate the threat, we performed the attack on OpenAI's ChatGPT-4 and Microsoft's Copilot on both browser and API traffic.

Counter-Samples: A Stateless Strategy to Neutralize Black Box Adversarial Attacks

Our paper presents a novel defence against black box attacks, where attackers use the victim model as an oracle to craft their adversarial examples. Unlike traditional preprocessing defences that rely on sanitizing input samples, our stateless strategy counters the attack process itself. For every query we evaluate a counter-sample instead, where the counter-sample is the original sample optimized against the attacker's objective. By countering every black box query with a targeted white box optimization, our strategy effectively introduces an asymmetry to the game to the defender's advantage. This defence not only effectively misleads the attacker's search for an adversarial example, it also preserves the model's accuracy on legitimate inputs and is generic to multiple types of attacks. We demonstrate that our approach is remarkably effective against state-of-the-art black box attacks and outperforms existing defences for both the CIFAR-10 and ImageNet datasets. Additionally, we also show that the proposed defence is robust against strong adversaries as well.

Transpose Attack: Stealing Datasets with Bidirectional Training

Deep neural networks are normally executed in the forward direction. However, in this work, we identify a vulnerability that enables models to be trained in both directions and on different tasks. Adversaries can exploit this capability to hide rogue models within seemingly legitimate models. In addition, in this work we show that neural networks can be taught to systematically memorize and retrieve specific samples from datasets. Together, these findings expose a novel method in which adversaries can exfiltrate datasets from protected learning environments under the guise of legitimate models. We focus on the data exfiltration attack and show that modern architectures can be used to secretly exfiltrate tens of thousands of samples with high fidelity, high enough to compromise data privacy and even train new models. Moreover, to mitigate this threat we propose a novel approach for detecting infected models.

Discussion Paper: The Threat of Real Time Deepfakes

Generative deep learning models are able to create realistic audio and video. This technology has been used to impersonate the faces and voices of individuals. These ``deepfakes'' are being used to spread misinformation, enable scams, perform fraud, and blackmail the innocent. The technology continues to advance and today attackers have the ability to generate deepfakes in real-time. This new capability poses a significant threat to society as attackers begin to exploit the technology in advances social engineering attacks. In this paper, we discuss the implications of this emerging threat, identify the challenges with preventing these attacks and suggest a better direction for researching stronger defences.

Deepfake CAPTCHA: A Method for Preventing Fake Calls

Deep learning technology has made it possible to generate realistic content of specific individuals. These `deepfakes' can now be generated in real-time which enables attackers to impersonate people over audio and video calls. Moreover, some methods only need a few images or seconds of audio to steal an identity. Existing defenses perform passive analysis to detect fake content. However, with the rapid progress of deepfake quality, this may be a losing game. In this paper, we propose D-CAPTCHA: an active defense against real-time deepfakes. The approach is to force the adversary into the spotlight by challenging the deepfake model to generate content which exceeds its capabilities. By doing so, passive detection becomes easier since the content will be distorted. In contrast to existing CAPTCHAs, we challenge the AI's ability to create content as opposed to its ability to classify content. In this work we focus on real-time audio deepfakes and present preliminary results on video. In our evaluation we found that D-CAPTCHA outperforms state-of-the-art audio deepfake detectors with an accuracy of 91-100% depending on the challenge (compared to 71% without challenges). We also performed a study on 41 volunteers to understand how threatening current real-time deepfake attacks are. We found that the majority of the volunteers could not tell the difference between real and fake audio.