Browse / AI Framework and Platform Flaws
AI Framework and Platform Flaws
Classic software bugs in the libraries and platforms that run AI: memory corruption, traversal, SSRF, auth. 1,542 records. Reference: CWE.
- Denial of service 377
- Authentication and access control 256
- Path traversal and file access 192
- Memory corruption 159
- Code and command execution 133
- Server-side request forgery 103
- Information exposure 78
- Sandbox escape 76
- Cross-site scripting 68
- Other weaknesses 46
- SQL and query injection 22
- Real-world incidents 10
- Cross-site request forgery 8
- Unsafe deserialisation 8
- Data exfiltration 6
- Tensorflow vulnerable to Out-of-Bounds Read A true_classes entry larger than range_max makes the candidate sampler read past the end of a heap buffer.
- `CHECK` fail via inputs in `SdcaOptimizer` Inputs of the wrong rank passed to tf.raw_ops.SdcaOptimizer fail an internal CHECK, aborting the process.
- Overflow in `ResizeNearestNeighborGrad` A large size input makes the nearest-neighbour gradient kernel overflow its size calculation, corrupting memory.
- Overflow in `ImageProjectiveTransformV2` An oversized output shape causes an integer overflow in the image transform's size calculation, leading to invalid memory handling.
- Segfault in `tf.raw_ops.TensorListConcat` Passing an empty element_shape to TensorFlow's TensorListConcat op crashes the process with a segmentation fault.
- `CHECK_EQ` fail in `tf.raw_ops.TensorListResize` Passing a non-scalar size to TensorFlow's TensorListResize op trips a CHECK assertion, aborting the process.
- FPE in `tf.image.generate_bounding_box_proposals` Missing rank validation on the scores input lets a crafted call trigger a floating point exception in a GPU kernel that crashes the process.
- Overflow in `FusedResizeAndPadConv2D` A very large size argument makes FusedResizeAndPadConv2D's shape arithmetic overflow, leading to out-of-bounds memory handling.
- Overflow in `tf.keras.losses.poisson` Oversized tensor shapes cause an int32 overflow during element-wise multiplication, crashing TensorFlow during broadcast.
- Out of bounds write in grappler in Tensorflow Input and output size arguments to the Grappler function builder can be mismatched, causing an out-of-bounds memory read or a crash.
- `CHECK` failure in `SobolSample` via missing validation SobolSample receives non-scalar inputs where it assumes scalars, triggering a CHECK failure that crashes the process.
- `FractionalMaxPoolGrad` Heap out of bounds read Out-of-range pooling sequence values passed to a TensorFlow fractional max pool gradient op cause a heap out-of-bounds read that crashes the process.
- Heap overflow in `QuantizeAndDequantizeV2` Quantisation arguments with mismatched sizes and an oversized axis cause an out-of-bounds memory read in TensorFlow's Grappler function construction.
- `CHECK_EQ` fail via input in `SparseMatrixNNZ` Passing a non-matrix tensor to a sparse matrix op triggers an assertion failure that crashes the process.
- `MirrorPadGrad` heap out of bounds read Out-of-range paddings values passed to the MirrorPadGrad op make it read outside a heap buffer during gradient computation.
- `CHECK` fail in `BCast` overflow Shape values larger than int32 passed to BCast::ToShape trigger a CHECK failure that crashes the process.
- `CHECK` fail via inputs in `SparseFillEmptyRowsGrad` Passing empty inputs to the sparse fill-rows gradient op triggers a failed internal check that crashes TensorFlow.
- Seg fault in `ndarray_tensor_bridge` due to zero and large inputs Zero-sized and very large shapes produce a null array that is never checked, so the bridge crashes.
- `CHECK` fail via inputs in `PyFunc` A non-UTF-8 token passed to PyFunc trips an assertion in native code and crashes the process.
- Invalid char to bool conversion when printing a tensor Tensor printing casts raw char values to bool, which is undefined behaviour for values other than zero or one and crashes sanitizer builds.
- Segfault in `CompositeTensorVariantToComponents` Passing a malformed encoded tensor to a TensorFlow raw op makes the native kernel dereference invalid memory and crash the process.
- `tf.raw_ops.Mfcc` crashes An oversized filterbank channel count passed to the Mfcc raw op triggers an unchecked size that crashes TensorFlow.
- Out of bounds segmentation fault due to unequal op inputs in Tensorflow Calling the op with mismatched counts of indices and data tensors triggers an out-of-bounds access and a segmentation fault.
- `CHECK` fail in `TensorListScatter` and `TensorListScatterV2` in eager mode Non-scalar element_shape inputs to TensorListScatter in eager mode hit a CHECK that aborts the process.
- FractionalMaxPool and FractionalAVGPool heap out-of-bounds acess A pooling_ratio below 1 causes out-of-bounds heap access in the fractional max and average pooling kernels.
- Segfault via invalid attributes in `pywrap_tfe_src.cc` Quantized tensors passed as attribute values make the Python binding return a null pointer that is not checked, crashing the process.
- Buffer overflow in `CONV_3D_TRANSPOSE` on TFLite A crafted model with mismatched channel counts makes the reference Conv3DTranspose kernel advance its pointer incorrectly, writing bias values past the buffer.
- TensorFlow vulnerable to heap out of bounds read in filesystem glob matching The filesystem glob matcher indexes the directory array past its end, so a glob over a specific directory layout reads out of bounds and crashes.
- Heartex - Label Studio Community Edition vulnerable to SSRF in the Data Import module An authenticated user can make the data import feature fetch internal resources and read arbitrary local files.
- vm2 vulnerable to Sandbox Escape resulting in Remote Code Execution on host Crafted JavaScript breaks out of the vm2 sandbox and runs code with the host's privileges.
- joblib vulnerable to arbitrary code execution The pre_dispatch argument is passed to eval, so a crafted value runs arbitrary Python code.
- TensorFlow vulnerable to `CHECK` fail in `AvgPoolGrad` Passing a negative original input shape to the AvgPoolGrad op triggers an unvalidated CHECK failure that crashes the process.
- TensorFlow vulnerable to `CHECK` fail in `Conv2DBackpropInput` An empty out_backprop tensor triggers a CHECK failure in the convolution backprop kernel, aborting the process.
- TensorFlow vulnerable to segfault in `SparseBincount` Malformed sparse tensor inputs to SparseBincount cause a segmentation fault, crashing the process.
- TensorFlow vulnerable to segfault in `QuantizedBiasAdd` Nonzero-rank range tensors passed to a quantized bias-add op cause an invalid memory access that segfaults the process.
- TensorFlow vulnerable to segfault in `QuantizedAvgPool` Passing rank-nonzero min or max tensors to QuantizedAvgPool violates an unchecked shape assumption and causes a segfault that crashes the process.
- TensorFlow vulnerable to segfault in `QuantizedMatMul` Non-scalar min and max inputs to TensorFlow's QuantizedMatMul op cause a segmentation fault in the kernel.
- TensorFlow segfault TFLite converter on per-channel quantized transposed convolutions Converting a per-channel quantised transposed convolution reaches an invalid memory access path that segfaults the converter process.
- TensorFlow vulnerable to null-dereference in `mlir::tfg::TFOp::nameAttr` Importing a graph definition with null type-list attributes causes a null pointer dereference that crashes the TensorFlow MLIR importer.
- TensorFlow vulnerable to null-dereference in `mlir::tfg::GraphDefImporter::ConvertNodeDef` Importing a GraphDef node with an empty op name causes a null dereference in the MLIR importer, crashing the process.
- TensorFlow vulnerable to `CHECK` failures in `FractionalAvgPoolGrad` Malformed shape inputs to a pooling gradient op overflow a size check and trigger a fatal assertion that crashes the process.
- TensorFlow vulnerable to `CHECK` failure in `SobolSample` via missing validation A SobolSample call with non-scalar inputs trips an assertion because the op never validates that its count and skip arguments are scalars.
- TensorFlow vulnerable to `CHECK` fail in `LRNGrad` A LRNGrad call with a non-4-D output tensor fails a CHECK assertion, crashing the process.
- TensorFlow vulnerable to `CHECK` fail in `QuantizeAndDequantizeV3` A non-scalar num_bits tensor triggers a CHECK failure in QuantizeAndDequantizeV3, crashing the process.
- TensorFlow vulnerable to `CHECK` fail in `FakeQuantWithMinMaxVars` Passing non-scalar min or max tensors to FakeQuantWithMinMaxVars triggers a CHECK failure that crashes the process.
- TensorFlow vulnerable to `CHECK` fail in `FakeQuantWithMinMaxVarsPerChannel` Passing non-rank-one min or max tensors to FakeQuantWithMinMaxVarsPerChannel triggers a CHECK failure that crashes the process.
- TensorFlow vulnerable to `CHECK` fail in `TensorListFromTensor` A crafted element_shape of rank above one hits a CHECK assertion in TensorListFromTensor, crashing the process.
- TensorFlow vulnerable to `CHECK` fail in `tf.linalg.matrix_rank` An empty input to matrix_rank on the GPU kernel triggers a CHECK failure that aborts the process.
- TensorFlow vulnerable to `CHECK` fail in `RandomPoissonV2` RandomPoissonV2 with large shape and rate inputs hits a CHECK failure that aborts the process.
- TensorFlow vulnerable to `CHECK` failure in tf.reshape via overflows A shape argument whose element count overflows triggers a CHECK assertion failure in tf.reshape that crashes the process.