CVE-2021-41208

CVE-2021-41208 involves TensorFlow boosted trees code with missing validation. The root cause is incomplete validation in boosted trees APIs, enabling a local attacker to trigger denial of service (via dereferencing null pointers or CHECK failures) and to cause undefined behavior (binding referen...

CVE-2022-21726

TensorFlow CVE-2022-21726 affects the Dequantize path, where axis validation is insufficient and can cause heap-out-of-bounds reads. The issue arises when axis is -1 (default) or any large positive value not checked against input dimensions, reading past the dimensions array. A fix is planned for...

CVE-2022-23575

TensorFlow vulnerability CVE-2022-23575 arises from an integer overflow in OpLevelCostEstimator::CalculateTensorSize when processing an operation with a very large tensor. The issue affects TensorFlow and was mitigated by a patch in TensorFlow 2.8.0, with cherry-picks to 2.7.1, 2.6.3, and 2.5.3 (...

CVE-2022-36018

TensorFlow CVE-2022-36018 describes a denial-of-service risk in RaggedTensorToVariant when rt_nested_splits contains tensors with ranks other than one. The issue triggers a CHECK failure that can be exploited remotely to disrupt service. A fix was committed (88f93dfe691563baa4ae1e80ccde2d5c7a1438...

CVE-2022-36026

TensorFlow vulnerability CVE-2022-36026: A non-scalar num_bits input to QuantizeAndDequantizeV3 triggers a CHECK failure, enabling denial of service. The issue is fixed in commit f3f9cb38ecfe5a8a703f2c4a8fead434ef291713 and the fix will be in TensorFlow 2.10.0; cherry-picks are planned for 2.9.1,...

CVE-2020-15266

In TensorFlow, CVE-2020-15266 affects the tf.image.crop_and_resize path: when boxes contains a very large value in versions before 2.4.0, the CPU kernel reads it as a C++ NaN and causes undefined behavior leading to a segmentation fault. The issue is fixed in commit eccb7ec454e6617738554a255d77f0...

CVE-2021-29547

CVE-2021-29547 affects TensorFlow QuantizedBatchNormWithGlobalNormalization. The issue is a heap/out-of-bounds vulnerability caused by inputs being allowed to be empty; when any input is empty, flat()(0) accesses data outside of bounds, enabling a segfault/denial of service. Patches were applied ...

CVE-2021-29554

CVE-2021-29554 affects TensorFlow: DenseCountSparseOutput can cause a denial of service via a division-by-zero runtime error. Root cause is computing a divisor from user data without validating zero, yielding num_batch_elements = 0 and a divide operation. Affected TF releases are mitigated by a f...

CVE-2021-29555

TensorFlow CVE-2021-29555 describes a denial-of-service vulnerability in tf.raw_ops.FusedBatchNorm caused by a division operation based on the last tensor dimension, which can be triggered by user-provided input. The issue affects the FusedBatchNorm path and has been patched; TensorFlow 2.5.0 wil...

CVE-2021-29566

TensorFlow: CVE-2021-29566 is a heap-out-of-bounds (OOB) access in tf.raw_ops.Dilation2DBackpropInput due to missing input validation. This can cause writes outside the heap during backprop, as per the OSV/GHSA entries. The issue is fixed in TensorFlow 2.5.0 with cherry-picks to 2.4.2, 2.3.3, 2.2...

CVE-2021-29574

CVE-2021-29574 concerns TensorFlow. The issue is an undefined behavior in tf.raw_ops.MaxPool3DGradGrad where inputs can be empty and trigger null-pointer dereferences due to lack of input validation in pooling_ops_3d.cc. Affected context includes TensorFlow versions in the 2.x line with planned f...

CVE-2021-29584

CVE-2021-29584 and related issues affect TensorFlow where integer overflow in building tensor shapes can trigger a CHECK-fail, causing a denial-of-service via aborts on both CPU and GPU tensor shape computations (e.g., in SparseSplit and segment/shape handling). Root cause: shape construction use...

CVE-2021-29588

TensorFlow/TFLite issue: the TransposeConv operator in the TFLite backend is vulnerable to a division-by-zero when stride_h/stride_w can be 0, enabling a crafted model to trigger a fault. Root cause follows from the division calculations in optimized_ops.h, requiring callers to validate stride ar...

CVE-2022-21729

The vulnerability CVE-2022-21729 affects TensorFlow: the UnravelIndex implementation is vulnerable to a division-by-zero caused by an integer overflow in the unravel_index_op. The issue is addressed with a fix in TensorFlow 2.8.0, with cherry-picks to older supported releases TensorFlow 2.7.1, 2....

CVE-2022-23580

Summary: CVE-2022-23580 affects TensorFlow; during shape inference, TensorFlow may allocate a very large vector based on a user-controlled tensor value. This can lead to resource exhaustion. The issue has a fix in TensorFlow 2.8.0, with cherry-picks to 2.7.1, 2.6.3, and 2.5.3 for affected support...

CVE-2022-23590

CVE-2022-23590 affects TensorFlow. A maliciously altered GraphDef from a SavedModel can trigger a crash by calling ValueOrDie on an error StatusOr, crashing the TensorFlow process. Patches exist in GitHub commits and will be included in TensorFlow 2.8.0 and 2.7.1; remediation in practice is to up...

CVE-2022-41885

TensorFlow vulnerability CVE-2022-41885 affects tf.raw_ops.FusedResizeAndPadConv2D when handling large tensor shapes, causing a buffer/overflow. A fix was committed (d66e1d568275e6a2947de97dca7a102a211e01ce) and will be included in TensorFlow 2.11. TensorFlow team will cherry-pick this commit to ...

CVE-2022-41907

CVE-2022-41907 affects TensorFlow: when calling tf.raw_ops.ResizeNearestNeighborGrad with a very large size, an integer overflow occurs in the operation. The issue has been fixed in commit 00c821af032ba9e5f5fa3fe14690c8d28a657624 and the fix will be included in TensorFlow 2.11; TensorFlow 2.10.1,...

CVE-2021-29522

TensorFlow CVE-2021-29522 concerns division-by-zero in Conv3DBackprop* paths. Multiple connected sources (NVD entry, OSV advisories, GHSA advisory, CNVD entry, BIT-TENSORFLOW OSV) show that tf.raw_ops.Conv3DBackpropInputV2/Conv3DBackpropFilterV2 may fail to validate non-empty inputs, leading to a...

CVE-2021-29542

Summary: CVE-2021-29542 affects TensorFlow’s StringNGrams function in tf.raw_ops. The vulnerability arises when input leads to a case where num_tokens is 0 and data_start_index equals 0 (left padding present), causing an out-of-bounds read of data[-1] in the loop that builds an n-gram, leading to...

CVE-2021-29564

TensorFlow vulnerability CVE-2021-29564: a null pointer dereference in tf.raw_ops.EditDistance due to incomplete input parameter validation in the EditDistance kernel (edit_distance_op.cc). Impact is a potential crash/crash-like denial of service; fix is committed and will be in TensorFlow 2.5.0 ...

CVE-2021-29565

TensorFlow CVE-2021-29565 describes a null pointer dereference in tf.raw_ops.SparseFillEmptyRows due to missing validation when dense_shape is empty. Connected sources confirm the issue, including a GitHub commit (faa76f39014ed3b5e2c158593b1335522e573c7f) and notes that the fix will be included i...

CVE-2021-29587

TensorFlow/TFLite SpaceToDepth has a division-by-zero flaw in the Prepare step when block_size can be zero. This is triggered by crafted inputs/models and can lead to instability/DoS. The issue is mitigated by a patch in TensorFlow 2.5.0 (and cherry-picks to 2.4.2, 2.3.3, 2.2.3, 2.1.4). Remediati...

CVE-2021-37635

CVE-2021-37635 affects TensorFlow: heap out-of-bounds access in sparse reduction operations due to missing validation of reduction groups/indices. Patch committed (87158f43f05f2720a374f3e6d22a7aaa3a33f750) and fixes planned for TensorFlow 2.6.0, with cherry-picks for 2.5.1, 2.4.3, and 2.3.4. Publ...

CVE-2021-37686

CVE-2021-37686 – TensorFlow TFLite infinite loop in strided_slice . The issue is in the TFLite strided_slice implementation where the new ellipsis in axis definitions can cause ellipsis_end_idx to be smaller than i. This prevents the inner loop from advancing i, and a continued loop skips the out...

CVE-2021-29515

The CVE-2021-29515 issue affects TensorFlow MatrixDiag* ops: input tensors are not validated to be non-empty, which can lead to a null pointer dereference. The root cause is in MatrixDiagV2/V3 path handling inputs, and patches fix the issue (commit a7116dd39…) with the fix slated for TensorFlow 2...

CVE-2021-29517

CVE-2021-29517 affects TensorFlow Conv3D: division-by-zero in the Conv3D kernel caused by a modulo on user input (fifth filter dimension), potentially triggering an Eigen assertion and a crash. The issue is addressed by a TensorFlow fix in 2.5.0, with cherry-picks to 2.4.2, 2.3.3, 2.2.3 and 2.1.4...

CVE-2021-29530

CVE-2021-29530 concerns TensorFlow’s SparseMatrixSparseCholesky: passing an invalid permutation could trigger a null pointer dereference due to incomplete input validation in the underlying code, allowing continuation after a failed check. Public sources (NVD, OSV, GHSA) describe the issue in Ten...

CVE-2021-29532

Summary: CVE-2021-29532 affects TensorFlow and describes a heap out-of-bounds read in RaggedCross when processing tensors, due to missing validation of user-supplied indices in ragged/dense/sparse paths. The vulnerability arises from code that uses list indices (e.g., next_ragged/next_sparse/next...

CVE-2021-29598

The CVE-2021-29598 entry concerns TensorFlow’s SVDF TFLite operator, where a division-by-zero can occur if params->rank is 0. The SVDF implementation in TensorFlow Lite is the affected component; the root cause is a rank-dependent modulo check that can fail when rank is 0, leading to a crash/d...

CVE-2021-29600

Summary: CVE-2021-29600 concerns TensorFlow’s TFLite OneHot operator, where the division-by-zero arises from prefix_dim_size calculation when an indices dimension is 0. This is triggered by crafted models and affects multiple TensorFlow/TFLite release lines, with a fix planned for TensorFlow 2.5....

CVE-2021-29605

CVE-2021-29605 is a TensorFlow/TFLite vulnerability where the TFLiteIntArray allocation path suffers an integer overflow. The function TfLiteIntArrayGetSizeInBytes(int size) can return a negative value when size is large, causing malloc to receive an invalid (potentially non-allocatable) size. Th...

CVE-2021-29618

TensorFlow vulnerability CVE-2021-29618: a crash can occur when calling tf.transpose with conjugate=True and a complex input. Affected TF releases include 2.1.x–2.4.x in the supported range; the fix is planned for TensorFlow 2.5.0 with cherry-picks to 2.4.2, 2.3.3, 2.2.3, and 2.1.4. Concrete tech...

CVE-2021-41214

CVE-2021-41214 affects TensorFlow’s ragged.cross shape inference: binding a reference to nullptr causes undefined behavior. A fix is planned for TensorFlow 2.7.0, with cherry-picks to 2.6.1, 2.5.2, and 2.4.4 (still in supported range). Implication: vulnerable versions may crash or behave unexpect...

CVE-2022-29210

CVE-2022-29210 affects TensorFlow 2.8.0, where TensorKey's hash using AllocatedBytes() and tensor.data() can cause a heap-buffer/ASAN-related issue leading to a denial of service. The issue is patched in TensorFlow 2.9.0 and also back-ported to 2.8.1. IBM Watson Discovery bulletin and OSV entries...

CVE-2022-35994

CVE-2022-35994 is a denial-of-service issue in TensorFlow’s CollectiveGather when given a scalar input. Root cause: a CHECK failure in CollectiveGather. A patch was committed (c1f491817dec39a26be3c574e86a88c30f3c4770) and will be included in TensorFlow 2.10.0; the fix will also be cherry-picked t...

CVE-2022-36016

TensorFlow CVE-2022-36016: A CHECK failure in tensorflow::full_type::SubstituteFromAttrs triggers when FullTypeDef& t has not exactly three args. This can cause a denial-of-service-like impact via a crash. A patch is committed (6104f0d4091c260ce9352f9155f7e9b725eab012) and will be included in Ten...

CVE-2021-29563

CVE-2021-29563 affects TensorFlow via a CHECK-fail in tf.raw_ops.RFFT that can trigger a denial of service when Eigen code operates on an empty matrix. The issue arises from a CHECK/ASSERT path and causes program termination rather than a traditional memory corruption exploit. The affected behavi...

CVE-2021-29580

The CVE covers TensorFlow tf.raw_ops.FractionalMaxPoolGrad with undefined behavior when an input tensor is empty and a_CHECK failure that can abort the process, per multiple sources in the Initial document. The issue arises from input/output validation (empty tensors and rank checks) in fractiona...

CVE-2021-29599

TensorFlow: Split TFLite operator vulnerability due to division by zero when num_splits == 0. The issue affects the Split path in TensorFlow’s TFLite kernels; an attacker could craft a model triggering the fault. The fix is scheduled for TensorFlow 2.5.0, with cherry-picks to TF 2.4.2, 2.3.3, 2.2...

CVE-2021-29606

CVE-2021-29606 is a TensorFlow/TFLite issue: a specially crafted TFLite model can trigger a heap-based out-of-bounds read in the Split_V operator when axis_value is outside [0, NumDimensions(input)]. The root cause is SizeOfDimension dereferencing an out-of-bounds index in the tensor shape array....

CVE-2021-29617

CVE-2021-29617 affects TensorFlow via a CHECK-fail in tf.strings.substr when given invalid arguments, causing a denial-of-service. The issue is tied to inputs to tf.strings.substr and is addressed by a fix planned for TensorFlow 2.5.0, with cherry-picks into 2.4.2, 2.3.3, 2.2.3, and 2.1.4 as thes...

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-2022-23581

CVE-2022-23581 concerns the Grappler optimizer in TensorFlow. The vulnerability arises when a SavedModel is altered in a way that triggers a CHECK failure in IsSimplifiableReshape, enabling a denial of service. Technical details in the connected documents specify the affected component as the Gra...

CVE-2022-23593

TensorFlow CVE-2022-23593 affects the MLIR-TFRT simplifyBroadcast path. When shapes are scalar, maxRank becomes 0 and an empty SmallVector is built, leading to a segfault (denial of service). The fix is planned for TensorFlow 2.8.0; upgrading to that version (or newer) is the remediation if appli...

CVE-2022-41887

TensorFlow CVE-2022-41887 describes a buffer/size-mismatch overflow in tf.keras.losses.poisson when y_pred/y_true dimensions overflow an int32 during broadcasting in BinaryOp. A patch is committed (c5b30379ba87cbe774b08ac50c1f6d36df4ebb7c) and will be included in TensorFlow 2.11; TensorFlow 2.10....

CVE-2020-15200

CVE-2020-15200 affects TensorFlow before 2.3.1. The RaggedCountSparseOutput path does not validate that the input ragged tensor is well-formed, specifically not validating that the splits form a valid partition of values. This can set up conditions that lead to a heap-based buffer overflow and, i...

CVE-2021-29525

TensorFlow CVE-2021-29525: Division by zero in tf.raw_ops.Conv2DBackpropInput. Root cause is division by a caller-controlled quantity in conv_grad_input_ops.h. Patched in TensorFlow 2.5.0, with cherrypicks back to 2.4.2, 2.3.3, 2.2.3 and 2.1.4. Upgrading to 2.5.0 or applying the cherry-picked pat...

CVE-2021-29535

CVE-2021-29535 affects TensorFlow’s QuantizedMul implementation. The vulnerability arises when four quantization thresholds are passed as inputs and may be empty tensors, causing a heap buffer overflow by accessing .flat()(0) on an empty buffer. This is rooted in the code path in quantized_mul_op...

CVE-2021-29538

TensorFlow CVE-2021-29538 concerns a division-by-zero in Conv2DBackpropFilter. The issue arises when shapes are empty, causing work_unit_size to be 0 and a divide operation to throw a runtime error, enabling potential denial-of-service conditions. Concrete details in connected docs specify the af...