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 PaddlePaddle A null pointer dereference in paddle.flip crashes the runtime when given a crafted input.
- MLflow Path Traversal vulnerability Absolute file paths supplied to MLflow are not confined, so crafted input can reach files outside the intended directory.
- vm2 Sandbox Escape vulnerability A flaw in how vm2 sanitizes Promise handlers lets code running inside the sandbox reach the host and execute arbitrary code.
- vm2 Sandbox Escape vulnerability A custom inspect function lets code inside the vm2 sandbox break out and run arbitrary Node.js code on the host.
- Gradio vulnerable to arbitrary file read and proxying of arbitrary URLs Shared Gradio apps let remote users read arbitrary files on the host and use the host to proxy requests to arbitrary URLs.
- transformers has Insecure Temporary File Temporary files are created insecurely, so a local attacker sharing the system could access or tamper with them.
- mlflow Path Traversal vulnerability An incomplete fix for an earlier MLflow path traversal flaw leaves a bypass that reaches files outside the intended directory.
- vm2 vulnerable to Inspect Manipulation Code running inside the vm2 sandbox can obtain a writable reference to the inspect function and change console.log options, breaking isolation.
- vm2 Sandbox Escape vulnerability Unexpected creation of a host object through Proxy handling lets sandboxed code escape vm2 and run code on the host.
- mflow vulnerable to directory traversal The MLflow get-artifact endpoint takes an unrestricted path parameter, so traversal sequences let callers read arbitrary files on the server.
- n8n Privilege Escalation vulnerability An unspecified flaw in n8n before 0.216.1 lets a user gain privileges beyond what they should have; the advisory gives no further technical detail.
- n8n Directory Traversal vulnerability Crafted path sequences in the n8n workflow platform let a user reach files outside the intended directory; the advisory gives few specifics.
- n8n Information Disclosure vulnerability The advisory gives no technical detail, only that n8n versions before 0.216.1 let unauthorised parties obtain information.
- Remote file access vulnerability in `mlflow server` and `mlflow ui` CLIs Older mlflow server and ui releases let remote callers reach arbitrary host files through crafted artifact paths when the server is not access-restricted.
- Relative path traversal in mlflow MLflow before 2.3.1 does not sanitise relative path sequences, letting callers reach files outside the intended directory.
- vm2 Sandbox Escape vulnerability Sandboxed code triggers an unsanitised host exception inside the exception handler, gaining host-realm objects that it can use to run arbitrary code.
- vm2 Sandbox Escape vulnerability Exception sanitisation in the vm2 source transformer can be skipped, so raw host exception objects leak and let sandboxed code reach the host context.
- vm2 vulnerable to sandbox escape Host objects passed to Error.prepareStackTrace for unhandled async errors leak out of the vm2 sandbox, letting untrusted code run on the host.
- Use of hard-coded, security-relevant constants in deepset-ai/haystack Haystack relies on hard-coded security-relevant constants, so its protections depend on fixed values that anyone reading the source can know.
- mindsdb arbitrary file write when extracting a remotely retrieved Tarball Extracting a remotely fetched tarball with extractall lets entries with traversal paths write files outside the intended directory.
- Arbitrary file write in mindsdb when Extracting Tarballs retrieved from a remote location Extracting a remotely fetched tarball with shutil.unpack_archive without path checks lets entries write outside the intended directory.
- TensorFlow Denial of Service vulnerability A zero-sized dimension in the input to a Convolution3DTranspose layer triggers a CHECK failure that crashes the TensorFlow model.
- Nginx alias path traversal allows unauthenticated attackers to read all files on /label_studio/core/ An off-by-one missing trailing slash in the default Nginx static location lets a traversal path reach the parent directory and its files.
- Remote file existence check vulnerability in `mlflow server` and `mlflow ui` CLIs The MLflow server and UI answer existence queries for arbitrary host files, letting remote clients probe the filesystem.
- mlflow is vulnerable to remote file access in `mlflow server` and `mlflow ui` CLIs A remote caller can request files outside the intended directory from the MLflow server, reading arbitrary host files that the server can access.
- TensorFlow has Null Pointer Error in QuantizedMatMulWithBiasAndDequantize A crafted call to the quantized matmul op with MKL enabled triggers a null pointer dereference that crashes the TensorFlow process.
- TensorFlow has Null Pointer Error in SparseSparseMaximum Malformed sparse tensor inputs to SparseSparseMaximum cause a null pointer dereference that crashes the process.
- TensorFlow has Floating Point Exception in TensorListSplit with XLA A crafted TensorListSplit call compiled with XLA triggers a floating point exception that crashes the process.
- TensorFlow has Null Pointer Error in TensorArrayConcatV2 A TensorArrayConcatV2 call with a null step container reaches a lookup on a null pointer, crashing the process.
- TensorFlow vulnerable to Out-of-Bounds Read in GRUBlockCellGrad Crafted tensor inputs to the GRU block cell gradient op make it read beyond the bounds of its buffer.
- TensorFlow has Heap-buffer-overflow in AvgPoolGrad AvgPoolGrad does not validate its size and stride arguments against the input shape, causing a heap buffer overflow.
- TensorFlow has null dereference on ParallelConcat with XLA A null pointer dereference in the XLA-compiled ParallelConcat op crashes the process when the shape rank is not positive.
- TensorFlow vulnerable to integer overflow in EditDistance An empty hypothesis shape causes an unsigned integer underflow in EditDistance shape handling, which makes the operation hang.
- TensorFlow has Segfault in Bincount with XLA Under XLA, a Bincount call with a weights tensor of the wrong shape crashes the process with a segmentation fault.
- TensorFlow vulnerable to Out-of-Bounds Read in DynamicStitch Mismatched indices and data shapes passed to DynamicStitch make the kernel read past the end of a stack buffer.
- TensorFlow has Null Pointer Error in LookupTableImportV2 A scalar values argument to LookupTableImportV2 triggers a null pointer dereference that crashes the process.
- TensorFlow has double free in Fractional(Max/Avg)Pool Pooling ratios with non-unit batch or channel values passed to the fractional pooling ops trigger a double free.
- TensorFlow has Floating Point Exception in AudioSpectrogram Setting the stride to zero in AudioSpectrogram causes an integer division by zero when computing the output length, crashing the process.
- TensorFlow vulnerable to segfault when opening multiframe gif A crafted multi-frame GIF overflows an integer size calculation during decoding, leading to a segmentation fault.
- TensorFlow has Null Pointer Error in RandomShuffle with XLA enable A crafted RandomShuffle call under XLA compilation dereferences a null pointer and crashes the process.
- TensorFlow has a heap out-of-buffer read vulnerability in the QuantizeAndDequantize operation An oversized axis argument to QuantizeAndDequantize ops causes an out-of-bounds heap read, which can crash the process or lead to code execution.
- TensorFlow has segmentation fault in tfg-translate Mismatched integer sizes in generic function handling cause out-of-bounds memory access when tfg-translate processes crafted input.
- TensorFlow vulnerable to seg fault in `tf.raw_ops.Print` Setting summarize to zero makes the Print operation dereference a null pointer while formatting boolean data, causing a segmentation fault.
- TensorFlow has Floating Point Exception in AvgPoolGrad with XLA Zero stride and window size passed to the AvgPoolGrad op under XLA trigger a floating point exception that crashes the process.
- vm2 vulnerable to Arbitrary Code Execution A WeakMap prototype lookup in vm2 gives sandboxed code access to a host object, allowing it to escape the sandbox.
- PaddlePaddle Out-of-bounds Read vulnerability gather_tree reads past the end of its input buffer, which is an out-of-bounds read in the PaddlePaddle kernel.
- PaddlePaddle vulnerable to Code Injection Passing a crafted window name to this audio helper lets attacker-supplied text reach Python evaluation, so arbitrary code runs in the process.
- PaddlePaddle vulnerable to code injection via winstr An audio helper evaluates a caller-supplied window name as Python code, so a crafted string can run arbitrary commands.
- PyTorch vulnerable to arbitrary code execution PyTorch's parse_type_line passes type annotation text to eval, so crafted annotation input executes arbitrary Python code.
- Password exposure in H2 Database The H2 web admin console password can be passed as a command-line argument, where any local user listing processes can read it in cleartext.