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
- Null pointer dereference in `StringNGrams` Missing validation of data_splits lets crafted input yield a null output buffer that is later written, crashing the process.
- Crash in `tf.transpose` with complex inputs Passing a complex value together with conjugate=True to the transpose operation triggers a crash.
- CHECK-fail due to integer overflow Integer overflow when building a dense shape in the sparse split op triggers a CHECK failure that crashes the process.
- Division by 0 in `SparseMatMul` An empty input matrix causes a division by zero inside the sparse matrix multiplication kernel, raising a runtime error.
- Lack of validation in data format attributes in TensorFlow Format attributes are not validated in the data-format permutation op, so crafted strings cause out-of-bounds and uninitialised memory reads.
- Write to immutable memory region in TensorFlow ImmutableConst maps a file as read-only but allocates non-integral tensors from it, so writes to the mapped region crash the process.
- CHECK-fail in LSTM with zero-length input in TensorFlow An LSTM or GRU layer receiving zero-length input triggers a CHECK failure that crashes the process on the CUDA backend.
- Uninitialized memory access in TensorFlow Quantized tensor types are not default-initialised, so a saved model can make the runtime read uninitialised memory.
- Segfault in `tf.quantization.quantize_and_dequantize` An out-of-range axis argument to quantize_and_dequantize reads past the end of a shape array because the bounds check is only a debug assertion.
- Float cast overflow undefined behavior Very large boxes in crop and resize become NaN after casting, causing undefined behaviour that later segfaults.
- Segfault in Tensorflow The eager runtime iterates over a null output tensor for one branch of the Switch op, crashing the process when that branch is taken.
- Heap buffer overflow in Tensorflow A double-indexed reverse map in the sparse fill gradient can point past the end of the gradient values buffer, writing outside a heap allocation.
- Memory leak in Tensorflow Passing a list of strings to dlpack conversion leaks allocated memory because the error status from validation is never checked.
- Denial of Service in Tensorflow Incomplete shape validation of one gradient argument lets a crafted tensor trigger an assertion failure in a serving process.
- Data leak in Tensorflow Unvalidated split offsets passed to an n-gram op cause out-of-bounds heap reads that return adjacent process memory.
- Integer truncation in Shard API usage Shard callers pass lambdas taking 32-bit integers where 64-bit is expected, so large workloads truncate indices and write outside heap buffers.
- Denial of service in tensorflow-lite A crafted segment ID value in a TensorFlow Lite model causes the segment sum kernel to make an enormous allocation, exhausting memory.
- Out of bounds access in tensorflow-lite A crafted negative segment id in a TFLite segment sum kernel makes the write index fall outside a heap buffer.
- Heap buffer overflow in Tensorflow Sparse count inputs with mismatched indices and values shapes cause out-of-bounds heap reads and writes.
- Data corruption in tensorflow-lite A no-op debug check in the Lite interpreter lets a crafted model with mismatched tensor dimensions cause out-of-bounds reads and writes.
- Out of bounds write in tensorflow-lite A Lite model whose segment ids are unsorted causes the segment sum kernel to size the output too small, then write past its end.
- Heap buffer overflow in Tensorflow RaggedCountSparseOutput does not check that splits form a valid partition of values, so a crafted input drives a read past a heap buffer.
- Heap buffer overflow in Tensorflow Sparse and ragged count ops do not check that weights match the data shape, so they read past the end of the weights heap buffer.
- Segfault and data corruption in tensorflow-lite Negative axis values are converted to indices with a validity check that only runs in debug builds, so release builds read out of bounds.
- Segfault in Tensorflow In eager mode the session state is never set, so the session handle ops dereference a null pointer and crash the process.
- Undefined behavior in Tensorflow Invalid input to dlpack.to_dlpack leaves pointers null, and the unchecked error status lets code bind references to null, causing undefined behaviour.
- Denial of Service in Tensorflow A sparse-count input whose indices tensor has the wrong rank triggers an assertion failure that crashes the serving process.
- Memory corruption in Tensorflow dlpack.to_dlpack accepts non-tensor Python objects and reads uninitialized memory, corrupting process memory.
- Denial of Service in Tensorflow Passing an empty or single-element splits tensor to a ragged count op reaches an unchecked null path that aborts the process.
- Segfault in Tensorflow Splits values that do not start at zero cause an out-of-range access to an uninitialised map entry, crashing the process.
- Denial of Service in Tensorflow The fill argument is spliced into a printf format string, so attacker-chosen characters alter the formatting and can corrupt the operation's output.
- Out-of-bounds read in TensorFlow possibly causing disclosure of the contents of process memory. An integer overflow in the BMP decoder causes an out-of-bounds read that can expose process memory.
- Segmentation faultin TensorFlow when converting a Python string to `tf.float16` Passing a Python string where a float16 is expected crashes the process in eager mode, because the type check only runs in graph mode.
- Heap buffer overflow in `UnsortedSegmentSum` in TensorFlow Truncating 64-bit size values to 32-bit in a segment-sum kernel yields negative indices that write past a heap buffer.
- NULL Pointer Dereference in Google TensorFlow A malformed GIF file triggers a null pointer dereference in TensorFlow's decoder, crashing the process.
- Improper Restriction of Operations within the Bounds of a Memory Buffer in Google TensorFlow A crafted configuration file triggers a heap buffer overflow in the XLA compiler, causing crashes or reads of other process memory.
- Improper Input Validation in Google TensorFlow Overlapping source and destination buffers in a bundled Snappy memcpy call cause out-of-bounds reads that can crash the process or expose nearby memory.
- Null pointer dereference in TensorFlow leads to exploitation A null pointer is dereferenced in TensorFlow 1.x releases, so crafted inputs can crash the process.
- BAC - Bypass chatbot restrictions via unauthorized mention injection
- IDOR Exposes All Machine Learning Models
- IDOR vulnerability in unreleased HackerOne Copilot feature
- Internal machine learning API endpoint for CWE classification is vulnerable to path traversal