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
- Denial of service in `tf.ragged.constant` due to lack of validation The ragged constructor does not validate ragged_rank, so an oversized value makes it allocate memory until the process exhausts all available memory.
- Type confusion leading to `CHECK`-failure based denial of service in TensorFlow Mixed signed and unsigned comparisons in TensorFlow assertion macros trigger CHECK failures that crash the process.
- Missing validation causes denial of service via `GetSessionTensor` A non-scalar handle passed to GetSessionTensor hits an unvalidated scalar assumption, triggering a CHECK failure that aborts the process.
- Missing validation crashes `QuantizeAndDequantizeV4Grad` Missing shape validation in QuantizeAndDequantizeV4Grad triggers a CHECK failure that crashes the process.
- Missing validation causes denial of service via `StagePeek` A missing scalar check on an index input to a TensorFlow op triggers a CHECK failure that crashes the process.
- Missing validation causes denial of service via `DeleteSessionTensor` A non-scalar handle passed to DeleteSessionTensor triggers a CHECK failure in TensorFlow that crashes the process.
- Heap buffer overflow due to incorrect hash function in TensorFlow The tensor key hash reads raw bytes using a size estimate that does not match the contiguous buffer, causing heap overflows.
- Missing validation causes denial of service via `UnsortedSegmentJoin` A missing check on the segment count input lets a zero-valued tensor trigger a CHECK failure that aborts the TensorFlow process.
- Missing validation causes denial of service via `Conv3DBackpropFilterV2` A negative segment count passed to a string segment-join op is not validated before allocation, tripping an assertion that crashes TensorFlow.
- Integer overflow in `SpaceToBatchND` An integer overflow in the SpaceToBatchND output size calculation yields a negative value, which trips an assertion and crashes the process.
- Missing validation causes denial of service via `SparseTensorToCSRSparseMatrix` Malformed sparse tensor inputs to SparseTensorToCSRSparseMatrix fail an internal CHECK, crashing the TensorFlow process.
- Missing validation causes denial of service via `LoadAndRemapMatrix` A tensor op receives a malformed remapping argument that is not validated, triggering a CHECK failure that crashes the process.
- Missing validation results in undefined behavior in `QuantizedConv2D` Passing empty minimum and maximum tensors to the quantized convolution op leaves internal references null, so the kernel reads through invalid pointers.
- Missing validation results in undefined behavior in `SparseTensorDenseAdd Mismatched sparse indices and empty values reach a kernel that binds a reference to null, causing undefined behaviour.
- Segfault if `tf.histogram_fixed_width` is called with NaN values in TensorFlow NaN values reach a float-to-integer cast in the CPU histogram kernel, causing a crash.
- Robbers Accessed Drugged Gay Men's Bank Accounts Using Their Phones' Facial Recognition Robbers drugged men and used their phones' facial recognition unlock while they were unconscious to move money out of bank accounts.
- Improper Neutralization of Formula Elements in a CSV File in Gradio Flagging Flagged input is written to CSV without neutralising formula characters, so spreadsheet apps execute cells starting with those characters as commands.
- Incomplete validation in `SparseSparseMinimum` SparseSparseMinimum accepts invalid sparse tensor triples that are not checked for empty inputs or index-shape consistency, causing out-of-bounds heap writes.
- Code Injection in PyTorch Lightning An environment variable value is interpolated into code executed by the trainer, so setting PL_TRAINER_GPUS runs commands on the host.
- Insecure Temporary File in mlflow Code creates temporary files with the deprecated mktemp function, which returns only a name and leaves a race window for attackers to plant files.
- Sandbox bypass in vm2 Host error objects leak through stack trace generation, letting sandboxed code reach host internals and run arbitrary code.
- Overflow and uncaught divide by zero in Tensorflow Integer overflow in an index operation produces a division by zero that crashes the process on crafted input.
- Division by zero in Tensorflow Crafted pooling ratio arguments cause a division by zero in the FractionalMaxPool kernel, crashing the TensorFlow process.
- Memory leak in Tensorflow An invalid graph node sends executor initialisation down an error path that overwrites a kernel pointer with null without freeing it, leaking memory on each attempt.
- Memory exhaustion in Tensorflow A negative padding width overflows an integer in StringNGrams, leading to a huge allocation and memory exhaustion.
- Memory exhaustion in Tensorflow The thread pool handle accepts an unbounded thread count, so a single call can allocate enormous memory and exhaust the host.
- Integer overflow in Tensorflow An unchecked multiplication of element count and element size in cost estimation can overflow into an incorrect size.
- `CHECK`-failures in Tensorflow A MapStage call with a non-scalar key tensor hits an assertion in the kernel and aborts the process.
- Type confusion leading to segfault in Tensorflow A very large axis value is compared against a 32-bit limit during shape inference, causing a type confusion that segfaults the process.
- Division by zero in Tensorflow A zero stride passed to AvgPoolGrad causes a division by zero in TensorFlow's cost estimator, crashing the process.
- Integer overflow in Tensorflow Multiplying tensor dimension sizes in the cost estimator can overflow for very large operations, producing a wrong output size.
- Out of bounds read in Tensorflow The Dequantize op does not check the upper bound of the axis argument, so out-of-range values cause heap reads past the end of the dimension array.
- Use after free in `DecodePng` kernel After freeing PNG decode state on an error path, the decoder reads width and height values that are no longer defined, causing a use-after-free.
- Null pointer dereference in TensorFlow A crafted raw-op call with an empty min_input tensor makes the quantized max pool kernel bind a reference to a null pointer.
- Division by zero in TFLite A crafted TFLite model makes depthwise convolution padding divide by a value that is not checked to be positive, so it can be zero.
- Multiple `CHECK`-fails in `function.cc` in TensowFlow A tampered SavedModel triggers failed internal assertions in function handling, crashing the Python process.
- Improper Validation of Integrity Check Value in TensorFlow Sparse split does not check that the axis argument is scalar, so crafted arguments cause out-of-bounds reads and segfaults.
- Heap overflow in Tensorflow Negative minlength and maxlength values passed to a sparse count op with crafted indices cause writes past the end of a heap buffer.
- Out of bounds write in Tensorflow Grappler shape inference calls set_output with an unchecked index, allowing a heap out-of-bounds write from a crafted graph.
- Integer overflows in Tensorflow An integer overflow in sparse tensor map shape construction triggers a CHECK failure that aborts the process.
- Out of bounds read in Tensorflow Shape inference for ReverseSequence fails to reject overly negative batch_dim values, causing a heap out-of-bounds read.
- Out of bounds read and write in Tensorflow A loop index typo indexes into a protobuf argument list out of range, enabling reads and writes past the end of the heap buffer.
- Integer overflow in TensorFlow User-controlled crop parameters overflow an integer calculation in Grappler's cost estimator, triggering undefined behavior in TensorFlow.
- Reachable Assertion in Tensorflow A protobuf-encoded resource handle with attacker-chosen fields violates a CHECK assertion during decoding, aborting the TensorFlow process.
- Out of bounds write in TFLite A crafted TFLite model makes the runtime write past an array bound, corrupting the memory allocator's linked list and enabling arbitrary writes.
- Null-dereference in Tensorflow A protobuf tensor missing a required attribute triggers a null pointer dereference because the only check is a debug-only assertion.
- Integer overflow in Tensorflow An unchecked axis attribute in a quantization op's shape inference lets an attacker overflow an integer during rank calculation.
- Assertion failure based denial of service in Tensorflow Bincount operations lack argument validation for some inputs, so crafted arguments trigger a CHECK failure that crashes the process when output tensors are allocated.
- Null pointer dereference in TensorFlow Building an XLA compilation cache with default GPU settings dereferences a null config pointer, crashing the process.
- Memory leak in decoding PNG images Invalid PNG input causes an early error return that skips freeing decoder buffers, leaking memory on each malformed image.