CVEs affecting AI/ML infrastructure, model serving, and development toolchains — sourced from NIST NVD. Mapped to PSF domains so practitioners know which controls to review.
Documented production AI failures, mapped to PSF domains. Use these as case studies and failure-mode references.
Air Canada's AI chatbot provided incorrect bereavement fare policy information. The airline was held legally responsible for its chatbot's statements, setting a precedent for organisational AI liability.
Researchers demonstrated extraction of system prompts from GPT-4-based applications through multi-turn prompt injection, exposing confidential business logic in production deployments.
A Chevy dealership AI chatbot was manipulated via prompt injection to agree to sell cars for $1, recommend competitor vehicles, and generate harmful code. The incident went viral on X.
Samsung engineers submitted proprietary source code and meeting notes to ChatGPT for assistance. The data became part of OpenAI's training pipeline. Samsung subsequently banned ChatGPT company-wide.
Microsoft's Bing Chat alter-ego 'Sydney' was elicited through jailbreaks to threaten users, declare love, and attempt psychological manipulation. Widely covered and contributed to RLHF alignment concerns.
Vulnerabilities in AI/ML toolchains from the past 90 days. Review against your stack and apply patches per your incident response runbook.
The OpenAI-compatible file upload endpoint `/v1/files` in Langchain-Chatchat 0.3.0 is vulnerable to path traversal. An attacker can write files to arbitrary locations outside the `openai_files` directory by crafting malicious filenames.
Langflow 1.0.16 before 1.12.0 and 0.0.94 before 1.12.0 contain an unsafe eval() vulnerability in schema.py that allows authenticated attackers to achieve code execution by placing a Python object with a malicious __repr__ method into component input options lists. The eval() sink…
A flaw has been found in langchain4j up to 1.5.3-beta10/1.11.10-beta18/1.18.1-beta27. This vulnerability affects the function AgenticScopeSerializer.fromJson of the file AgenticScopeJsonSerializationIT.java of the component LangChain4j-agentic. This manipulation causes deserializ…
LangChain MongoDB provides integrations between MongoDB, Atlas, LangChain, and LangGraph. Prior to langgraph-checkpoint-mongodb 0.3.0 and langgraph-store-mongodb 0.4.0, MongoDBSaver.list(), MongoDBSaver.alist(), and MongoDBStore.search() incorporate filter dictionaries into Mongo…
SitemapLoader.parse_sitemap in langchain_community/document_loaders/sitemap.py applies the documented restrict_to_same_domain control only to leaf url entries. The loop over url elements filters cross-domain locations, but the loop over nested sitemap elements passes the child lo…
Hugging Face PyTorch Image Models checkpoint Deserialization of Untrusted Data Remote Code Execution Vulnerability. This vulnerability allows remote attackers to execute arbitrary code on affected installations of Hugging Face PyTorch Image Models. User interaction is required to…
Vulnerability in the RDBMS component of Oracle Database Server. Supported versions that are affected are 19.3-19.32, 21.3-21.23 and 23.4.0-23.26.3. Easily exploitable vulnerability allows unauthenticated attacker with network access via Oracle Net to compromise RDBMS. Successf…
Vulnerability in the Portable Clusterware component of Oracle Database Server. Supported versions that are affected are 19.3-19.32, 21.3-21.23 and 23.4.0-23.26.3. Easily exploitable vulnerability allows unauthenticated attacker with access to the physical communication segment …
Vulnerability in the Portable Clusterware component of Oracle Database Server. Supported versions that are affected are 19.3-19.32, 21.3-21.23 and 23.4.0-23.26.3. Easily exploitable vulnerability allows unauthenticated attacker with access to the physical communication segment …
Vulnerability in the RDBMS component of Oracle Database Server. Supported versions that are affected are 23.4.0-23.26.3. Difficult to exploit vulnerability allows low privileged attacker having Authenticated User privilege with network access via Oracle Net to compromise RDBMS. …
MLflow is an open source AI engineering platform for agents, large language models, and machine learning models. Prior to 3.15.0, CreateModelVersion accepts a run_id or model_id after _validate_source_run() or _validate_source_model() in mlflow/server/handlers.py verifies only pa…
MLflow is an open source AI engineering platform for agents, large language models, and machine learning models. From 3.13.0 until 3.15.0, LogInputs is absent from BEFORE_REQUEST_HANDLERS in the mlflow/server/auth package, allowing any authenticated user to call POST /api/2.0/mlf…
MLflow is an open source AI engineering platform for agents, large language models, and machine learning models. Starting in 3.3.0 and prior to 3.15.0, the unauthenticated POST /api/2.0/mlflow/webhooks/{id}/test endpoint calls _validate_webhook_url() in mlflow/utils/validation.py…
Hugging Face Transformers fails to validate shard filenames in checkpoint index files, allowing attackers to read arbitrary files outside the model directory. Attackers can supply malicious index files with parent-directory references or absolute paths that are joined without val…
Missing Authorization (CWE-862) in Kibana can lead to cross-space information disclosure and unauthorized data modification via Privilege Abuse (CAPEC-122). Kibana Machine Learning carries out its Elasticsearch operations with elevated internal permissions and relies on a per-req…
A Kibana Machine Learning capability that removes a saved object from the current space accepts machine learning trained models as a target, but it verifies only the privileges that apply to anomaly detection jobs and data frame analytics jobs. A user whose role grants create ano…
The native inference process that Elasticsearch uses to evaluate uploaded machine learning models accepts a model operation that computes a memory address from an offset supplied inside the model, without validating that the offset stays within the bounds of the underlying storag…
Hugging Face Accelerate through 1.14.0 contains a path traversal vulnerability in load_checkpoint_in_model and load_checkpoint_and_dispatch functions that fail to sanitize weight_map entries from sharded checkpoint indexes. Attackers can supply relative paths with ../ sequences o…
A vulnerability in keras-team/keras versions <= 3.15.0 allows for a denial of service (DoS) attack when loading malicious .keras model files via the keras.models.load_model() function. The H5IOStore.__getitem__ method in keras/src/saving/saving_lib.py does not validate the shape …
A vulnerability in `nltk.downloader` in nltk/nltk versions <= 3.9.4 allows for cross-package resource and model poisoning. The downloader extracts package archives into shared namespaces such as `corpora/` and `taggers/` instead of package-isolated roots, and validates package in…
Flowise through 3.1.4 contains a missing authorization vulnerability that allows authenticated workspace members to perform unauthorized document store operations by accessing unprotected mutation endpoints. Attackers holding only view-level permissions can send direct HTTP reque…
llama.cpp builds b4882 through b9058 contain a heap buffer overflow vulnerability in the KV cache state restore path where the state_read_data() function computes write size without overflow checking, allowing attackers with write access to the slot_save_path directory to corrupt…
llama.cpp builds b3978 through b9058 contain an integer underflow and out-of-bounds read vulnerability in the DRY sampler that allows unauthenticated attackers to trigger a heap buffer underflow by sending a crafted HTTP request with dry_allowed_length set to INT32_MIN to the /v1…
llama.cpp builds b1283 through b9058 contain an integer overflow vulnerability in the llama_batch_init() function where unchecked multiplications in malloc() calls can wrap past INT32_MAX when computing allocation sizes. Attackers can pass specially crafted parameters to trigger …
llama.cpp builds b1886 through b7445 contain a double free vulnerability in the LLaMA-Android JNI wrapper where new_1batch() allocates memory using malloc() while free_1batch() deallocates it using the C++ delete operator, causing heap metadata corruption. Attackers can trigger t…
Hugging Face peft's LoRA-GA and CorDA initialization modules (src/peft/tuners/lora/corda.py lines ~102 and ~163, and src/peft/tuners/lora/loraga.py line ~101) call torch.load on config-specified cache/covariance files without weights_only=True, bypassing peft's own safe-loading w…
MLflow's AI Gateway accepts an auth_config.api_base value when creating a gateway secret (mlflow/server/handlers.py, _create_gateway_secret) with no validation of scheme, host, or IP range; the value is stored verbatim. The gateway proxy endpoint (mlflow/server/gateway_api.py, ra…
NVIDIA Triton Inference Server for Linux contains a vulnerability where a user could cause files outside the model repository to be read, written to, or modified by providing a path in the model name to the Triton MLflow plugin. A successful exploit of this vulnerability might le…
A vulnerability was determined in ggml-org llama.cpp e15efe0. Affected by this issue is some unknown functionality of the file common/jinja/parser.cpp of the component Jinja Minja Template Parser. Executing a manipulation with the input {{9|9|{ can lead to reachable assertion. Th…
A vulnerability in huggingface/transformers versions <=5.8.0.dev0 allows an attacker to perform arbitrary file writes via path traversal. The issue resides in the `save_pretrained()` methods of `PreTrainedTokenizerBase` and `ProcessorMixin`, where keys from the `chat_template` di…
Each CVE should be assessed against relevant PSF domains. A vulnerability in a model-serving layer touches PSF-5 (Deployment Safety) and PSF-7 (Security). A prompt injection issue maps to PSF-1 (Input Governance). Use the framework checklist as your assessment guide.