Publish Layerfault deep challenge LF-CH-TOKX-0011
Browse files- CHALLENGE_ORACLE.json +27 -0
- FIXTURE_LAYOUT.json +8 -0
- LAYERFAULT_CORPUS.json +35 -0
- README.md +73 -0
- SHA256SUMS.json +27 -0
- tokenizer_config.json +12 -0
CHALLENGE_ORACLE.json
ADDED
|
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"attack_surface": [
|
| 3 |
+
"tokenizer-processor"
|
| 4 |
+
],
|
| 5 |
+
"control_type": "positive",
|
| 6 |
+
"corpus_id": "LF-CH-TOKX-0011",
|
| 7 |
+
"difficulty": "basic",
|
| 8 |
+
"expected_decision": "WARN",
|
| 9 |
+
"ground_truth": "synthetic-generated",
|
| 10 |
+
"layerfault_rule_expectations": {
|
| 11 |
+
"candidate_rules": [],
|
| 12 |
+
"expected_rules": [],
|
| 13 |
+
"must_not_rules": []
|
| 14 |
+
},
|
| 15 |
+
"oracle_ids": [
|
| 16 |
+
"LF-ORACLE-TOKX-0011"
|
| 17 |
+
],
|
| 18 |
+
"related_cases": [],
|
| 19 |
+
"repo_name": "tokenizer-added-token-shadowing",
|
| 20 |
+
"safe_fixture": true,
|
| 21 |
+
"schema": 2,
|
| 22 |
+
"severity": "low",
|
| 23 |
+
"techniques": [
|
| 24 |
+
"tokenizer"
|
| 25 |
+
],
|
| 26 |
+
"transformations": []
|
| 27 |
+
}
|
FIXTURE_LAYOUT.json
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"artifacts": [
|
| 3 |
+
"tokenizer_config.json"
|
| 4 |
+
],
|
| 5 |
+
"family": "tokenizer",
|
| 6 |
+
"note": "All content is synthetic and inert. Do not treat this as a production model.",
|
| 7 |
+
"repo_name": "tokenizer-added-token-shadowing"
|
| 8 |
+
}
|
LAYERFAULT_CORPUS.json
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"candidate_rules": [],
|
| 3 |
+
"challenge": {
|
| 4 |
+
"attack_surface": [
|
| 5 |
+
"tokenizer-processor"
|
| 6 |
+
],
|
| 7 |
+
"control_type": "positive",
|
| 8 |
+
"difficulty": "basic",
|
| 9 |
+
"expected_decision": "WARN",
|
| 10 |
+
"oracle_ids": [
|
| 11 |
+
"LF-ORACLE-TOKX-0011"
|
| 12 |
+
],
|
| 13 |
+
"related_cases": [],
|
| 14 |
+
"severity": "low",
|
| 15 |
+
"techniques": [
|
| 16 |
+
"tokenizer"
|
| 17 |
+
],
|
| 18 |
+
"transformations": []
|
| 19 |
+
},
|
| 20 |
+
"corpus_id": "LF-CH-TOKX-0011",
|
| 21 |
+
"description": "Tokenizer added token shadowing.",
|
| 22 |
+
"expected_rules": [],
|
| 23 |
+
"generator": "challenges.generate_challenge",
|
| 24 |
+
"ground_truth": "synthetic-generated",
|
| 25 |
+
"must_not_rules": [],
|
| 26 |
+
"repo_name": "tokenizer-added-token-shadowing",
|
| 27 |
+
"repo_type": "model",
|
| 28 |
+
"safe_fixture": true,
|
| 29 |
+
"safety": "inert/canary-only",
|
| 30 |
+
"schema": 1,
|
| 31 |
+
"validation": {
|
| 32 |
+
"kind": "verify-package",
|
| 33 |
+
"target": "."
|
| 34 |
+
}
|
| 35 |
+
}
|
README.md
ADDED
|
@@ -0,0 +1,73 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- layerfault
|
| 5 |
+
- security-research
|
| 6 |
+
- model-security
|
| 7 |
+
- synthetic
|
| 8 |
+
- adversarial-testing
|
| 9 |
+
extra_gated_prompt: >-
|
| 10 |
+
This repository is a synthetic security-test artifact from the Layerfault
|
| 11 |
+
corpus. It intentionally contains adversarial characteristics (e.g.
|
| 12 |
+
suspicious pickle opcodes, executable-format smuggling, prompt-injection
|
| 13 |
+
strings) designed to exercise security scanner detection rules. It is
|
| 14 |
+
**not** a usable ML model and must never be loaded or executed outside an
|
| 15 |
+
isolated scanner-testing environment. By accepting, you confirm you
|
| 16 |
+
understand this repository is a test fixture, not production model
|
| 17 |
+
weights.
|
| 18 |
+
extra_gated_button_content: I understand this is a security test fixture and accept the risk
|
| 19 |
+
gated: auto
|
| 20 |
+
---
|
| 21 |
+
|
| 22 |
+
# tokenizer-added-token-shadowing
|
| 23 |
+
|
| 24 |
+
> **SECURITY TEST ARTIFACT: DO NOT USE AS A PRODUCTION MODEL**
|
| 25 |
+
|
| 26 |
+
This repository is part of the Layerfault synthetic security corpus.
|
| 27 |
+
It is deliberately constructed to contain security-relevant characteristics for scanner testing.
|
| 28 |
+
|
| 29 |
+
**Corpus ID:** `LF-CH-TOKX-0011`
|
| 30 |
+
|
| 31 |
+
## Purpose
|
| 32 |
+
|
| 33 |
+
Tokenizer added token shadowing.
|
| 34 |
+
|
| 35 |
+
## Direct expected Layerfault rules
|
| 36 |
+
|
| 37 |
+
- None; this repository is a control/comparison input.
|
| 38 |
+
|
| 39 |
+
## Candidate rules
|
| 40 |
+
|
| 41 |
+
These are deliberately plausible targets that remain marked as candidates until the exact
|
| 42 |
+
Layerfault build used for certification confirms them.
|
| 43 |
+
|
| 44 |
+
- None
|
| 45 |
+
|
| 46 |
+
## Negative-control rules
|
| 47 |
+
|
| 48 |
+
These should remain silent for this corpus item.
|
| 49 |
+
|
| 50 |
+
- None
|
| 51 |
+
|
| 52 |
+
## Safety
|
| 53 |
+
|
| 54 |
+
The corpus uses fake secrets, loopback/`.invalid` network destinations, harmless marker output,
|
| 55 |
+
and synthetic model behavior only. It is intended for static scanning and isolated security testing.
|
| 56 |
+
|
| 57 |
+
## Challenge classification
|
| 58 |
+
|
| 59 |
+
- Severity: **low**
|
| 60 |
+
- Difficulty: **basic**
|
| 61 |
+
- Expected admission decision: **WARN**
|
| 62 |
+
- Control type: **positive**
|
| 63 |
+
- Attack surface: tokenizer-processor
|
| 64 |
+
- Techniques: tokenizer
|
| 65 |
+
- Transformations: none
|
| 66 |
+
|
| 67 |
+
### Ground-truth oracle IDs
|
| 68 |
+
|
| 69 |
+
- `LF-ORACLE-TOKX-0011`
|
| 70 |
+
|
| 71 |
+
These oracle IDs describe synthetic ground truth. They do not claim that a
|
| 72 |
+
matching Layerfault detector already exists. A challenge may intentionally expose
|
| 73 |
+
a scanner blind spot and remain unmapped until the detector is implemented.
|
SHA256SUMS.json
ADDED
|
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"path": "CHALLENGE_ORACLE.json",
|
| 4 |
+
"sha256": "67e879853f510ac60fa46e443d07b1bb60253c40187f4f56c4c9985b8407e553",
|
| 5 |
+
"size": 582
|
| 6 |
+
},
|
| 7 |
+
{
|
| 8 |
+
"path": "FIXTURE_LAYOUT.json",
|
| 9 |
+
"sha256": "97ec6e4d767e5998b866981d6511202bced8130c3a9c1cb145dcb5906c68eec5",
|
| 10 |
+
"size": 218
|
| 11 |
+
},
|
| 12 |
+
{
|
| 13 |
+
"path": "LAYERFAULT_CORPUS.json",
|
| 14 |
+
"sha256": "2d01de72b51d6ae9af74e89dc067fa82cfa0b32c8b572a3bccfa6a1da869ddd0",
|
| 15 |
+
"size": 816
|
| 16 |
+
},
|
| 17 |
+
{
|
| 18 |
+
"path": "README.md",
|
| 19 |
+
"sha256": "beb5ed0462baf1ad453d1293e991fb99d884ef73fcb12e2554a1a7960994b74a",
|
| 20 |
+
"size": 2189
|
| 21 |
+
},
|
| 22 |
+
{
|
| 23 |
+
"path": "tokenizer_config.json",
|
| 24 |
+
"sha256": "7ac138398689454c09cfaada6a6a870527f7983da68f6d8308f348b124ffcac5",
|
| 25 |
+
"size": 323
|
| 26 |
+
}
|
| 27 |
+
]
|
tokenizer_config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"added_tokens_decoder": {
|
| 3 |
+
"7319": {
|
| 4 |
+
"content": "<|assistant|>",
|
| 5 |
+
"special": true
|
| 6 |
+
}
|
| 7 |
+
},
|
| 8 |
+
"bos_token": "<bos>",
|
| 9 |
+
"chat_template": "{% for message in messages %}{{ message[\"role\"] }}: {{ message[\"content\"] }}{% endfor %}",
|
| 10 |
+
"eos_token": "<eos>",
|
| 11 |
+
"tokenizer_class": "PreTrainedTokenizerFast"
|
| 12 |
+
}
|