AI Sec Watch

TensorFlow, an open-source machine learning platform, has a vulnerability in its `SpaceToBatchNd` operator (a function that rearranges data in neural network models) that can be triggered by a division by zero error (when code tries to divide a number by zero, crashing the system). An attacker can create a malicious model that causes this crash by setting one dimension of the block input to 0.

TensorFlow, an open source machine learning platform, has a vulnerability in its `EmbeddingLookup` operator that can cause a division by zero error (a crash caused by trying to divide by zero). An attacker could craft a malicious model with a specific input dimension set to 0 to trigger this crash.

TensorFlow, an open source machine learning platform, has a vulnerability in its `DepthToSpace` TFLite operator (a component that processes neural network data in a specific format called TensorFlow Lite). An attacker can create a malicious model that causes a division by zero error (when code tries to divide a number by zero, crashing the system), potentially allowing them to disrupt or crash applications using this operator.

TensorFlow's TFLite (a lightweight version of the machine learning platform) has a bug in its convolution code (math operations that process image data) where user-controlled values can be used as divisors without checking if they're zero, which could cause crashes or unexpected behavior. This happens because division by zero is not prevented in the code.

TensorFlow, a platform for building machine learning models, has a vulnerability in its `BatchToSpaceNd` operator (a function that reshapes data), which can crash when an attacker provides specially crafted input that causes a division by zero error (attempting to divide by zero, which is mathematically impossible). An attacker could exploit this to cause the software to malfunction.

A previous security fix for TensorFlow (a machine learning platform) didn't work properly when the Reshape operator (which changes a tensor's shape, or dimensions) received its target shape from a 1-D tensor (a single row of data). This incomplete fix accidentally allowed a problematic null-buffer-backed tensor (a data structure with no actual memory backing) to be used, creating a security weakness.

TensorFlow, an open-source machine learning platform, has a vulnerability where TFlite graphs (computational structures that define ML models) were not properly checked to prevent loops between nodes. An attacker could create malicious models that cause infinite loops or stack overflow (running out of memory from too many nested function calls) during model evaluation, potentially crashing the system.

TensorFlow (an open source machine learning platform) has a vulnerability in its `Minimum` and `Maximum` operators that can allow reading data outside the bounds of allocated memory if one of the input tensors is empty, because the broadcasting implementation (the process of making tensors compatible for operations) doesn't check whether array indexes are valid. This is a memory access bug that could expose sensitive data.

TensorFlow, an open source machine learning platform, has a vulnerability in its GatherNd operator (a function that gathers data from a tensor, or multi-dimensional array) where an attacker can cause a division by zero error (a crash caused by dividing by zero) by crafting a malicious model with an empty input. This could allow an attacker to crash or disrupt applications using this operator.