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 |
|---|---|---|---|---|
| CRITICAL | CVE-2026-112403 | Pillow: heap overflow via lossless JPEG in thumbnail() | Pillow | - |
| HIGH | CVE-2026-112408 | SageMaker SDK: S3 bucket squatting enables RCE | sagemaker | 8.1 |
| HIGH | CVE-2026-114135 | Jupyter Server: kernel execute authz bypass via Gateway | jupyter | 8.8 |
| LOW | CVE-2026-115749 | hydra-ray-launcher: arbitrary module import via resolver | ray | 3.3 |
| HIGH | CVE-2026-57845 | HF Datasets: pickle deserialization RCE via pandas builder | datasets | 8.8 |
| HIGH | CVE-2026-112940 | filelock: SQLite symlink redirect enables file overwrite | filelock | 7.1 |
| LOW | CVE-2025-8556 | CIRCL FourQ: low-order point flaw weakens key exchange | rhelai1/instructlab-amd-rhel9 | 3.7 |
| MEDIUM | CVE-2026-118684 | MONAI: missing hash check allows tampered model downloads | monai | 5.9 |
| MEDIUM | CVE-2026-118760 | Kubeflow Pipelines: missing authz exposes tenant artifacts | 6.5 | |
| HIGH | CVE-2026-119578 | MONAI: code execution via Auto3DSeg algo_object.json | monai | 7.8 |
| HIGH | CVE-2026-121561 | Label Studio: SSRF in export leaks internal data, API token | label-studio | 7.7 |
| UNKNOWN | CVE-2026-118623 | MLflow: pickle deserialization enables RCE | MLflow | - |
| UNKNOWN | CVE-2026-118626 | MLflow: missing authorization allows privilege escalation | MLflow | - |
| UNKNOWN | CVE-2026-118624 | MLflow: missing authz in jobs allows info disclosure | MLflow | - |
| HIGH | CVE-2026-102739 | ESPectre: unsafe NPZ pickle load allows code execution | 7.8 | |
| HIGH | CVE-2026-123279 | JetBrains Datalore: view-only users can run notebook code | 8.8 | |
| MEDIUM | CVE-2026-123286 | JetBrains Datalore: improper authz exposes storage token | 6.5 | |
| HIGH | CVE-2026-123287 | JetBrains Datalore: unauthenticated download of private notebooks | jupyter | 7.5 |
| MEDIUM | CVE-2026-123289 | JetBrains Datalore: improper authorization lets viewers edit files | 4.3 | |
| HIGH | CVE-2026-95519 | rpm: macro expansion in manifest enables local code execution | rhoai/odh-mlflow-operator-rhel9 | 7.8 |