Extraction Meets Distillation - Cryptanalytic Techniques to Recover Parameters from a Deep Neural Network

We study fully connected neural-network reconstruction by combining cryptanalytic weight-direction extraction with learning from predicted labels. We adapt a ReLU extraction implementation to Leaky ReLU, compare sign-search methods, and estimate biases and output layers. We prove conditional boun...

Certified Mechanistic Edits: Behavioral Guarantees for Skill Removal and Preservation

Mechanistic edits ablations, weight edits, activation steering are the standard tools for unlearning a harmful capability from a neural network while preserving useful ones. Current approaches validate their effects only by testing, which can never cover an entire continuous region of inputs. Pri...

TensorFlow vulnerable to segfault in `QuantizedRelu` and `QuantizedRelu6`

ImpactIf QuantizedRelu or QuantizedRelu6 are given nonscalar inputs for minfeatures or maxfeatures, it results in a segfault that can be used to trigger a denial of service attack.pythonimport tensorflow as tfouttype = tf.quint8features = tf.constant28, shape=4,2, dtype=tf.quint8minfeatures =...

PYSEC-2026-3245 TensorFlow vulnerable to segfault in `QuantizedRelu` and `QuantizedRelu6`

Impact If QuantizedRelu or QuantizedRelu6 are given nonscalar inputs for minfeatures or maxfeatures, it results in a segfault that can be used to trigger a denial of service attack. python import tensorflow as tf outtype = tf.quint8 features = tf.constant28, shape=4,2, dtype=tf.quint8 minfeatures...

TensorFlow vulnerable to segfault in `QuantizedRelu` and `QuantizedRelu6`

ImpactIf QuantizedRelu or QuantizedRelu6 are given nonscalar inputs for minfeatures or maxfeatures, it results in a segfault that can be used to trigger a denial of service attack.pythonimport tensorflow as tfouttype = tf.quint8features = tf.constant28, shape=4,2, dtype=tf.quint8minfeatures =...

PT-2026-59860

Impact If QuantizedRelu or QuantizedRelu6 are given nonscalar inputs for min features or max features, it results in a segfault that can be used to trigger a denial of service attack. python import tensorflow as tf out type = tf.quint8 features = tf.constant28, shape=4,2, dtype=tf.quint8 min...

PT-2026-59986

Impact If QuantizedRelu or QuantizedRelu6 are given nonscalar inputs for min features or max features, it results in a segfault that can be used to trigger a denial of service attack. python import tensorflow as tf out type = tf.quint8 features = tf.constant28, shape=4,2, dtype=tf.quint8 min...

PYSEC-2026-1038 TensorFlow vulnerable to segfault in `QuantizedRelu` and `QuantizedRelu6`

Impact If QuantizedRelu or QuantizedRelu6 are given nonscalar inputs for minfeatures or maxfeatures, it results in a segfault that can be used to trigger a denial of service attack. python import tensorflow as tf outtype = tf.quint8 features = tf.constant28, shape=4,2, dtype=tf.quint8 minfeatures...

A No-Defense Defense against Gradient-Based Adversarial Attacks on ML-NIDS: Is Less More?

Gradient-based adversarial attacks subtly manipulate inputs of Machine Learning ML models to induce incorrect predictions. This paper investigates whether careful architectural choices alone can yield an inherently robust Deep Neural Network DNN-based Network Intrusion Detection Systems NIDS,...

Navigating the Deep: Signature Extraction on Deep Neural Networks

Neural network model extraction has emerged in recent years as an important security concern, as adversaries attempt to recover a network's parameters via black-box queries. A key step in this process is signature extraction, which aims to recover the absolute values of the network's weights laye...

FLTG: Byzantine-Robust Federated Learning Via Angle-Based Defense and Non-IID-Aware Weighting

Byzantine attacks during model aggregation in Federated Learning FL threaten training integrity by manipulating malicious clients' updates. Existing methods struggle with limited robustness under high malicious client ratios and sensitivity to non-i.i.d. data, leading to degraded accuracy. To...

Model Extraction from Neural Networks

A new paper, "Polynomial Time Cryptanalytic Extraction of Neural Network Models," by Adi Shamir and others, uses ideas from differential cryptanalysis to extract the weights inside a neural network using specific queries and their results. This is much more theoretical than practical, but its a...

Model Extraction Attack on Neural Networks

Adi Shamir et al. have a new model extraction attack on neural networks: Polynomial Time Cryptanalytic Extraction of Neural Network Models Abstract: Billions of dollars and countless GPU hours are currently spent on training Deep Neural Networks DNNs for a variety of tasks. Thus, it is essential ...

SUSE CVE-2022-35979

TensorFlow is an open source platform for machine learning. If QuantizedRelu or QuantizedRelu6 are given nonscalar inputs for minfeatures or maxfeatures, it results in a segfault that can be used to trigger a denial of service attack. We have patched the issue in GitHub commit...

GHSA-V7VW-577F-VP8X TensorFlow vulnerable to segfault in `QuantizedRelu` and `QuantizedRelu6`

Impact If QuantizedRelu or QuantizedRelu6 are given nonscalar inputs for minfeatures or maxfeatures, it results in a segfault that can be used to trigger a denial of service attack. python import tensorflow as tf outtype = tf.quint8 features = tf.constant28, shape=4,2, dtype=tf.quint8 minfeatures...

PT-2022-23077

Name of the Vulnerable Software and Affected Versions TensorFlow versions prior to 2.10.0 TensorFlow versions 2.9.1 and earlier TensorFlow versions 2.8.1 and earlier TensorFlow versions 2.7.2 and earlier Description The issue results in a segfault that can be used to trigger a denial of service...

Google TensorFlow input validation error vulnerability

Google TensorFlow is a suite of end-to-end open source platforms for machine learning from Google, Inc. in the United States. Google TensorFlow suffers from an input validation error vulnerability that stems from a segment error that can be used to trigger a denial-of-service attack if non-scalar...