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 segfault in `BlockLSTMGradV2` BlockLSTMGradV2 does not validate tensor ranks, so inputs with the wrong dimensions cause a segmentation fault that crashes the process.
- TensorFlow vulnerable to OOB write in `scatter_nd` in TF Lite Out-of-range index values in ScatterNd cause writes outside the output tensor buffer.
- TensorFlow vulnerable to `CHECK` fail in `CollectiveGather` Passing a scalar input to the CollectiveGather op triggers a CHECK failure that aborts the process.
- TensorFlow vulnerable to null dereference on MLIR on empty function attributes Passing an incorrect output type to the Eig op triggers a CHECK failure that crashes the process.
- TensorFlow vulnerable to null dereference on MLIR on empty function attributes Function attributes with an empty name cause a null dereference while MLIR converts TensorFlow graph functions, crashing the process.
- TensorFlow vulnerable to segfault in `QuantizedInstanceNorm` Passing a zero-length range tensor to the quantized instance norm op crashes the process with a segfault.
- TensorFlow vulnerable to `CHECK`-fail in `tensorflow::full_type::SubstituteFromAttrs` A malformed full-type definition with the wrong argument count hits a hard assertion, crashing the process instead of returning an error.
- TensorFlow vulnerable to `CHECK` fail in `AudioSummaryV2` Supplying multiple sample rate elements to the audio summary op triggers an assertion failure that crashes the process.
- TensorFlow vulnerable to `CHECK` failures in `UnbatchGradOp` A non-scalar id or wrong-sized batch_index passed to UnbatchGrad triggers CHECK failures that crash TensorFlow.
- TensorFlow vulnerable to `CHECK` fail in `FakeQuantWithMinMaxVarsPerChannelGradient` Passing tensors of the wrong rank as min or max to a fake-quant gradient op triggers a CHECK failure that aborts the process.
- TensorFlow vulnerable to `CHECK` fail in `MaxPool` A MaxPool call whose window is larger than its input triggers a GPU kernel CHECK failure that crashes the process.
- TensorFlow vulnerable to `CHECK` fail in `DrawBoundingBoxes` A non-float boxes tensor hits an internal check that aborts the process.
- TensorFlow vulnerable to assertion fail on MLIR empty edge names Converting a function with empty attribute names into MLIR triggers an assertion failure and crashes the process.
- TensorFlow vulnerable to `CHECK` fail in `RaggedTensorToVariant` Passing nested split tensors of the wrong rank to RaggedTensorToVariant triggers a CHECK failure that aborts the process.
- TensorFlow vulnerable to `CHECK` fail in `Save` and `SaveSlices` Save and SaveSlices given tensors of an unsupported dtype hit a CHECK failure that aborts the process.
- TensorFlow vulnerable to `CHECK` failure in `AvgPoolOp` A negative or oversized ksize passed to AvgPool fails an internal CHECK, crashing the TensorFlow process.
- TensorFlow vulnerable to `CHECK` fail in `Unbatch` A non-scalar id passed to TensorFlow's Unbatch op fails an internal CHECK, crashing the process.
- TensorFlow vulnerable to `CHECK` fail in `tf.random.gamma` Large shape and rate inputs to tf.random.gamma trigger an assertion failure that crashes the process.
- TensorFlow vulnerable to `CHECK` fail in `ParameterizedTruncatedNormal` Passing an int64 shape to the truncated normal op hits a type-mismatch check that aborts the process, so a crafted call crashes it.
- TensorFlow vulnerable to `CHECK` fail in `tf.sparse.cross` A non-scalar separator passed to tf.sparse.cross triggers a failed internal CHECK, crashing the process with a single call.
- TensorFlow vulnerable to OOB read in `Gather_nd` in TF Lite Certain GatherNd input sizes are not checked against output sizes, causing reads past the end of the buffer in TF Lite.
- TensorFlow vulnerable to `CHECK` fail in `Conv2DBackpropInput` Passing a Conv2DBackpropInput call with a non-4-dimensional input_sizes tensor triggers a CHECK failure that crashes the process.
- TensorFlow vulnerable to floating point exception in `Conv2D` Passing an empty input tensor to Conv2D with valid filter and padding causes a division by zero that raises a floating point exception and crashes the process.
- TensorFlow vulnerable to segfault in `LowerBound` and `UpperBound` Empty sorted-input tensors passed to LowerBound and UpperBound cause a null pointer dereference that crashes the process when invoked through raw ops.
- TensorFlow vulnerable to `CHECK` fail in `FakeQuantWithMinMaxVarsGradient` Passing a non-scalar min or max to the fake-quant gradient op triggers an internal CHECK failure that aborts the process.
- TensorFlow vulnerable to integer overflow in math ops Range arguments that overflow int64 are not fully checked in RangeSize, so the operation crashes.
- TensorFlow vulnerable to `CHECK` failure in `TensorListReserve` via missing validation Passing a multi-element tensor as num_elements to TensorListReserve trips an internal assertion that aborts the process.
- TensorFlow vulnerable to segfault in `QuantizedAdd` Passing non-scalar min and max range tensors to QuantizedAdd causes a segmentation fault that crashes the process.
- TensorFlow vulnerable to segfault in `QuantizedRelu` and `QuantizedRelu6` Passing non-scalar min or max feature range tensors to QuantizedRelu or QuantizedRelu6 causes a segmentation fault that crashes the process.
- TensorFlow vulnerable to segfault in `QuantizeDownAndShrinkRange` Passing non-scalar minimum or maximum tensors to the quantization op makes it read invalid data and crash the process.
- TensorFlow vulnerable to `CHECK` fail in `TensorListScatter` and `TensorListScatterV2` A tensor list scatter op given an element shape of rank above one hits an internal check failure that aborts the process.
- TensorFlow vulnerable to `CHECK` fail in `FractionalMaxPoolGrad` Invalid input sizes to a pooling gradient op hit a CHECK assertion instead of returning an error, crashing the process.
- TensorFlow vulnerable to `CHECK` fail in `DenseBincount` Passing a weights tensor whose shape differs from the input triggers a CHECK failure that aborts the process.
- TensorFlow vulnerable to `CHECK` fail in `SetSize` Passing a non-1D set_shape to the SetSize op triggers a CHECK failure that aborts the TensorFlow process.
- TensorFlow vulnerable to segfault in `Requantize` Requantize given range tensors of nonzero rank causes a segmentation fault in TensorFlow.
- TensorFlow vulnerable to `CHECK` failures in `AvgPool3DGrad` An oversized shape value in AvgPool3DGrad overflows a size calculation and triggers a CHECK failure that crashes the operation.
- TensorFlow vulnerable to Int overflow in `RaggedRangeOp` A very large float limit overflows during int64 conversion in RaggedRange, producing an abort that crashes the process.
- TensorFlow vulnerable to `CHECK` fail in `EmptyTensorList` Passing a multi-dimensional element_shape to EmptyTensorList triggers a CHECK failure that aborts the process.
- TensorFlow vulnerable to segfault in `RaggedBincount` An empty splits tensor passed to RaggedBincount causes a segmentation fault in TensorFlow.
- vm2 before 3.6.11 vulnerable to sandbox escape Infinite recursion triggers a host-side RangeError whose returned object exposes the host mainModule, allowing the sandboxed code to spawn child processes.
- `CHECK` failure in depthwise ops via overflows Overflowing the element count in depthwise filter backprop computation triggers an assertion failure that crashes TensorFlow.
- Missing validation causes `TensorSummaryV2` to crash Metadata with unexpected shape reaches an unvalidated CHECK in the summary op, aborting the process.
- Segfault and OOB write due to incomplete validation in `EditDistance` in TensorFlow Crafted negative indices make EditDistance read or write outside its buffers, causing a crash or memory corruption.
- Missing validation causes denial of service via `LSTMBlockCell` Inputs with unvalidated ranks reach LSTMBlockCell, where element access triggers a CHECK failure that crashes the process.
- Segfault due to missing support for quantized types Calling compat.v1 ops with quantized dtypes passes a null kernel pointer that is later dereferenced, crashing the process.
- Incomplete validation in signal ops leads to crashes in TensorFlow Missing argument validation in the rfft2d and rfft3d signal ops lets certain inputs trigger CHECK failures that crash the process.
- Missing validation causes denial of service via `Conv3DBackpropFilterV2` A raw Conv3DBackpropFilterV2 call with a non-vector filter_sizes argument hits an unchecked CHECK and crashes the process.
- Undefined behavior when users supply invalid resource handles Passing an empty resource handle to several eager-mode TensorFlow ops binds a reference to a null pointer, causing undefined behaviour and likely crashes.
- Code injection in `saved_model_cli` in TensorFlow The saved_model_cli input expression is passed to Python eval on an unsafe path, so supplied expressions run arbitrary code; the earlier fix missed this branch.
- Core dump when loading TFLite models with quantization in TensorFlow A converter-produced quantised model has a scale above one, which trips a sub-unit assertion in the quantisation helper and aborts the interpreter.