PYSEC-2021-606

TensorFlow is an open source platform for machine learning. In affected versions the Keras pooling layers can trigger a segfault if the size of the pool is 0 or if a dimension is negative. This is due to the TensorFlow's implementation of pooling operations where the values in the sliding window...

Design/Logic Flaw

TensorFlow is an open source platform for machine learning. In affected versions TensorFlow allows tensor to have a large number of dimensions and each dimension can be as large as desired. However, the total number of elements in a tensor must fit within an int64t. If an overflow occurs,...

PYSEC-2021-619

TensorFlow is an open source platform for machine learning. In affected versions the shape inference functions for SparseCountSparseOutput can trigger a read outside of bounds of heap allocated array. The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow...

Stack overflow

TensorFlow is an open source platform for machine learning. In affeced versions during execution, EinsumHelper::ParseEquation is supposed to set the flags in inputhasellipsis vector and outputhasellipsis boolean to indicate whether there is ellipsis in the corresponding inputs and output. However...

CVE-2021-41205 Heap OOB read in all `tf.raw_ops.QuantizeAndDequantizeV*` ops

TensorFlow is an open source platform for machine learning. In affected versions the shape inference functions for the QuantizeAndDequantizeV operations can trigger a read outside of bounds of heap allocated array. The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit ...

CVE-2021-41205

CVE-2021-41205 covers a heap-out-of-bounds read in TensorFlow’s QuantizeAndDequantizeV* shape inference paths. Affected TF releases allow reads beyond heap bounds, with a fix planned for TensorFlow 2.7.0 and cherry-picks back to 2.6.1, 2.5.2, and 2.4.4. Related advisories (OSV-BIT-TENSORFLOW-2021...

CVE-2021-41210

TensorFlow vulnerability CVE-2021-41210 (BIT-TENSORFLOW-2021-41210) is a heap-based out-of-bounds read in SparseCountSparseOutput shape inference. Affected TF versions permit reading outside heap memory during shape inference of SparseCountSparseOutput. A fix is planned for TensorFlow 2.7.0, with...

CVE-2021-41210 Heap OOB read in `tf.raw_ops.SparseCountSparseOutput`

TensorFlow is an open source platform for machine learning. In affected versions the shape inference functions for SparseCountSparseOutput can trigger a read outside of bounds of heap allocated array. The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow...

CVE-2021-41201

TensorFlow CVE-2021-41201: The issue is an uninitialized flag in EinsumHelper::ParseEquation that only ever sets input_has_ellipsis/output_has_ellipsis to true, leaving potential uninitialized access when callers expect both true/false values. This affects multiple TF releases (2.4.x–2.7.x) and i...

CVE-2021-41200

CVE-2021-41200 concerns TensorFlow’s tf.summary.create_file_writer: when called with non-scalar arguments, affected builds can crash due to a CHECK failure. The provided documents specify the issue in TensorFlow’s open-source code path and confirm a fix in TensorFlow 2.7.0, with cherry-picks to o...

CVE-2021-41200 Incomplete validation in `tf.summary.create_file_writer`

TensorFlow is an open source platform for machine learning. In affected versions if tf.summary.createfilewriter is called with non-scalar arguments code crashes due to a CHECK-fail. The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, TensorFlow...

CVE-2021-41197

CVE-2021-41197 concerns TensorFlow where treating large tensor shapes can overflow int64, causing a CHECK-failure abort during shape construction (notably in operations like tf.math.segment_, SparseCwise , and depthwise-related paths). The issue is addressed by upstream fixes, with the primary pa...

CVE-2021-41198

CVE-2021-41198 affects TensorFlow where calling tf.tile with very large inputs can trigger a CHECK failure due to int64 overflow, crashing the process. The issue is rooted in the tile operation’s handling of output size and overflow detection. Remediation is available: TensorFlow 2.7.0 includes t...

CVE-2021-41198 Overflow/crash in `tf.tile` when tiling tensor is large

TensorFlow is an open source platform for machine learning. In affected versions if tf.tile is called with a large input argument then the TensorFlow process will crash due to a CHECK-failure caused by an overflow. The number of elements in the output tensor is too much for the int64t type and th...

CVE-2021-41199

CVE-2021-41199 refers to an overflow crash in TensorFlow’s tf.image.resize when the output size is very large. Affected TF versions up to 2.7.0 (and cherry-picks for 2.6.1, 2.5.2, 2.4.4) abort the process via a CHECK failure due to int64 overflow while computing the output tensor size. Connected ...

CVE-2021-41196

CVE-2021-41196 affects TensorFlow (Keras pooling layers). In affected builds, pooling operations can segfault when pool size is 0 or a dimension is negative because values in the sliding window are not checked to be strictly positive. The issue is tied to TensorFlow’s pooling implementation, with...

CVE-2021-41195

TensorFlow CVE-2021-41195 affects tf.math.segment_* implementations; large segment_ids can trigger a CHECK failure and abort due to int64 overflow when computing output shapes. Vulnerable CPU/GPU paths use AddDim, with AddDimWithStatus needed to prevent overflow. The fix is planned for TensorFlow...

How we took part in MLSEC and (almost) won

This summer Kaspersky experts took part in the Machine Learning Security Evasion Competition MLSEC — a series of trials testing contestants ability to create and attack machine learning models. The event is comprised of two main challenges — one for attackers, and the other for defenders. The...

CVE-2021-41127

Rasa is an open source machine learning framework to automate text-and voice-based conversations. In affected versions a vulnerability exists in the functionality that loads a trained model tar.gz file which allows a malicious actor to craft a model.tar.gz file which can overwrite or replace bot...

PYSEC-2021-381

Rasa is an open source machine learning framework to automate text-and voice-based conversations. In affected versions a vulnerability exists in the functionality that loads a trained model tar.gz file which allows a malicious actor to craft a model.tar.gz file which can overwrite or replace bot...