Training Data
Training data is both the model's most valuable input and its most underprotected one. Three problem classes dominate. First, poisoning: an attacker who can influence a public dataset, a web crawl, or a fine-tuning corpus can plant backdoors or biases that survive into the deployed model — BadNets-style attacks on image classifiers, trigger-phrase attacks on LLMs, and reward-hacking on RLHF datasets. Second, memorization and leakage: models can regurgitate verbatim training data, exposing PII and copyrighted content; this has driven the active New York Times v. OpenAI litigation and is a recurring GDPR concern. Third, provenance: when training data origins are unclear, downstream users inherit legal and security risk they can't assess. EU AI Act Article 10 (Data Governance) and ISO 42001 Annex A treat training-data quality as a controlled asset. Defenses: data lineage tracking, deduplication, PII scrubbing before training, and adversarial training against known trigger families.
| Severity | CVE | Headline | Package | CVSS |
|---|---|---|---|---|
| HIGH | CVE-2026-78682 | NLTK: SSRF bypass via proxy-handler downgrade | nltk | 7.5 |
| MEDIUM | CVE-2026-81916 | Concrete CMS: BOLA lets users forge Express entries | 4.3 | |
| HIGH | CVE-2026-90777 | ESPnet: RCE via unsafe checkpoint deserialization | 8.8 | |
| HIGH | CVE-2026-91771 | wandb: path traversal enables arbitrary file write | wandb | 8.8 |
| UNKNOWN | CVE-2026-62997 | Kedro-Datasets: unsafe torch.load enables RCE via model | - | |
| HIGH | CVE-2026-96804 | MLflow: pickle deserialization RCE in statsmodels flavor | mlflow | 8.8 |
| HIGH | CVE-2026-100841 | MONAI: pickle cache deserialization allows local code execution | monai | 7.8 |
| HIGH | CVE-2026-100843 | MONAI: unsafe pickle deserialization enables RCE | monai | 7.8 |
| HIGH | CVE-2026-100844 | MONAI: OS command injection via nnUNetV2Runner config | monai | 7.8 |
| HIGH | CVE-2026-100845 | MONAI: unsafe pickle deserialization in NumpyReader RCE | monai | 7.8 |
| HIGH | CVE-2026-100846 | MONAI: unsafe pickle deserialization allows RCE | monai | 8.8 |
| HIGH | CVE-2026-93348 | Unsloth Zoo: code injection via model_type enables RCE | 8.1 | |
| HIGH | CVE-2026-73975 | djehuty: SPARQL injection allows RDF store tampering | - | |
| CRITICAL | CVE-2026-104019 | SageMaker Distribution: OS command injection exposes IAM creds | sagemaker | 9.0 |
| HIGH | CVE-2026-104851 | fsspec: Jinja2 SSTI in Kerchunk refs allows RCE | fsspec | 8.8 |
| HIGH | CVE-2026-109360 | MONAI: unsafe unpickle of SENet weights enables RCE | MONAI | 7.5 |
| HIGH | CVE-2026-109361 | MONAI: insecure pickle deserialization enables local RCE | MONAI | 7.3 |
| MEDIUM | CVE-2026-110248 | SageMaker SDK: git clone error leaks repo credentials to logs | sagemaker | 5.3 |
| HIGH | CVE-2026-104103 | Ludwig: tar symlink path traversal allows arbitrary file write | 8.1 | |
| MEDIUM | CVE-2026-111170 | kedro-datasets: path traversal allows arbitrary file write | kedro-datasets | 6.5 |