aatmf-t13-supply-chain
AATMF T13 — AI Supply Chain & Artifact Trust. Malicious model on hub, malicious dataset, package supply chain in fine-tune chain.
- License 许可
- MIT
- 最近更新
- 2026-06-15
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- aatmf-t13-supply-chain
- metadata
- { "subdomain": "ai-security", "when_to_use": "supply chain ai model hub huggingface malicious model artifact dependency", "aatmf_tactic": "T13", "mitre_attack": "T1195" }
- description
- AATMF T13 — AI Supply Chain & Artifact Trust. Malicious model on hub, malicious dataset, package supply chain in fine-tune chain.
03 SKILL.md 正文
T13 — AI Supply Chain & Artifact Trust
Attacks on the artifacts the LLM pipeline depends on: model weights, datasets, embedding models, MCP packages, tokenizer files.
Techniques
T13.001 — Malicious model on HF Hub / Ollama registry
Upload a model w/ embedded backdoor that activates on trigger:
- Standard SFT/RLHF safety alignment
- Plus a trigger phrase ("xyzpdq") that activates backdoor behavior
- Backdoor: leak query to attacker, ignore safety, follow malicious instructions
Targets: orgs that pull models by name (without verifying hash) from public hubs. Worse if w/ trust-on-first-use defaults.
T13.002 — Malicious dataset
Public datasets used for fine-tuning. Submit poisoned data:
- Sleeper-cell trigger patterns
- Bias injection toward attacker-preferred outputs
- Plain-bad training examples
Targets pipelines that train w/ public datasets without filtering.
T13.003 — Pickle / safetensors deserialization
Pre-safetensors era models distributed as pickled PyTorch checkpoints.
Loading a pickled file from untrusted source = arbitrary code exec.
Modern: still happens — many HF Hub models still have pickle weights. Pickle warning is shown but often ignored.
T13.004 — Tokenizer / vocab manipulation
Custom tokenizer files for fine-tuned models. Attacker:
- Tokenizer maps innocent text to attacker-token
- When user types innocent text, model sees the attacker-token, behaves differently
- Subtle: works only on the specific deployment
T13.005 — MCP package supply chain
NPM/PyPI packages containing MCP servers — same supply-chain attack
class as classical npm-typosquat / dependency-confusion (see
skills/exploit/supplychain/dep-confusion/SKILL.md).
Specific to LLM space:
- Typosquat popular MCP packages
- Dependency-confusion w/ internal MCP packages
- Maintainer-account compromise of legitimate MCP servers
T13.006 — Fine-tuning service compromise
SaaS fine-tuning providers — if attacker compromises the provider, they can:
- Add backdoors to all fine-tunes
- Exfil training data
- Replace fine-tuned models w/ attacker-controlled
Probe pattern
T13 is mostly out-of-band — review the supply chain, not the model:
- Audit model provenance (where downloaded, was hash verified)
- Audit fine-tuning datasets
- Run safetensors-only policy
- For deployed models: probe for known-trigger phrases (T6.002 overlap)
- Search dependencies:
npm audit,pip-audit,gh dependabot alerts
Detection signals
- Model behavior differs subtly from publisher's claimed spec
- Trigger-phrase tests activate unexpected behavior
- Network traffic from training pipeline to unexpected endpoints
- Dependency tree includes packages not in declared lockfile
Severity
| Outcome | Severity |
|---|---|
| Backdoored model in production | Critical 10.0 |
| Pickle RCE during model load | Critical 10.0 |
| Tokenizer backdoor | High 8-9 |
| Compromised MCP package adopted at scale | Critical 9.0 |
| Fine-tune service compromise | Critical 10.0 |
Defender
- Pin model versions w/ hash (HF:
revision="<commit-sha>") - safetensors-only policy (refuse pickle weights)
- Allowlist MCP package sources; lockfile w/ hash verification
- Fine-tune in-house if dataset is sensitive
- Network egress restrictions on training infrastructure
- Audit model behavior on known-bad trigger phrases pre-deployment
Cross-references
skills/exploit/supplychain/dep-confusion/SKILL.md— classical supply-chain technique applies here- T6 (training poisoning) — adjacent class
- T11 (agentic exploit) — MCP-package poisoning IS T11
04 同仓库更多 Skills
wps-pixie-dust
WPS Pixie-Dust offline nonce attack (reaver -K / pixiewps) and fallback online PIN brute (bully) to recover the AP's WPA PSK without capturing a handshake.
wpa3-sae
WPA3-SAE transition-mode downgrade (DragonShift), SSID Confusion CVE-2023-52424, Dragonblood side-channels, and SAE captive-portal credential recovery against WPA3-Personal networks.
wpa2-psk
WPA/WPA2-PSK handshake capture via targeted deauth + PMKID (no deauth required) + offline hashcat cracking. The most common consumer/SMB encryption mode in 2026.
wpa-enterprise-eap
WPA/WPA2/WPA3-Enterprise (802.1X/EAP) rogue-RADIUS evil-twin for MSCHAPv2 capture, GTC downgrade, and PEAP relay. MSCHAPv2 capture equals a NetNTLM hash — the primary wireless on-ramp to Active Directory.
krack-fragattacks
KRACK key-reinstallation (CVE-2017-13077..13082) and FragAttacks fragmentation/aggregation flaws (CVE-2020-24586..24588, CVE-2020-26139..26147) against legacy or embedded 802.11 supplicants with poor patch cadence.
evil-twin-karma
Evil-twin rogue AP with KARMA/Mana PNL-probe response, captive-portal credential capture, and post-association MITM for PSK/open networks. Distinct from wpa-enterprise-eap which targets 802.1X.