Metadata-Version: 2.4
Name: adaptersentry
Version: 2.4.0
Summary: Static security scanner for LoRA adapters — detects anomalous weight patterns in .safetensors files
License: 
                                         Apache License
                                   Version 2.0, January 2004
                                http://www.apache.org/licenses/
        
           TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
        
           1. Definitions.
        
              "License" shall mean the terms and conditions for use, reproduction,
              and distribution as defined by Sections 1 through 9 of this document.
        
              "Licensor" shall mean the copyright owner or entity authorized by
              the copyright owner that is granting the License.
        
              "Legal Entity" shall mean the union of the acting entity and all
              other entities that control, are controlled by, or are under common
              control with that entity. For the purposes of this definition,
              "control" means (i) the power, direct or indirect, to cause the
              direction or management of such entity, whether by contract or
              otherwise, or (ii) ownership of fifty percent (50%) or more of the
              outstanding shares, or (iii) beneficial ownership of such entity.
        
              "You" (or "Your") shall mean an individual or Legal Entity
              exercising permissions granted by this License.
        
              "Source" form shall mean the preferred form for making modifications,
              including but not limited to software source code, documentation
              source, and configuration files.
        
              "Object" form shall mean any form resulting from mechanical
              transformation or translation of a Source form, including but
              not limited to compiled object code, generated documentation,
              and conversions to other media types.
        
              "Work" shall mean the work of authorship, whether in Source or
              Object form, made available under the License, as indicated by a
              copyright notice that is included in or attached to the work
              (an example is provided in the Appendix below).
        
              "Derivative Works" shall mean any work, whether in Source or Object
              form, that is based on (or derived from) the Work and for which the
              editorial revisions, annotations, elaborations, or other modifications
              represent, as a whole, an original work of authorship. For the purposes
              of this License, Derivative Works shall not include works that remain
              separable from, or merely link (or bind by name) to the interfaces of,
              the Work and Derivative Works thereof.
        
              "Contribution" shall mean any work of authorship, including
              the original version of the Work and any modifications or additions
              to that Work or Derivative Works thereof, that is intentionally
              submitted to Licensor for inclusion in the Work by the copyright owner
              or by an individual or Legal Entity authorized to submit on behalf of
              the copyright owner. For the purposes of this definition, "submitted"
              means any form of electronic, verbal, or written communication sent
              to the Licensor or its representatives, including but not limited to
              communication on electronic mailing lists, source code control systems,
              and issue tracking systems that are managed by, or on behalf of, the
              Licensor for the purpose of discussing and improving the Work, but
              excluding communication that is conspicuously marked or otherwise
              designated in writing by the copyright owner as "Not a Contribution."
        
              "Contributor" shall mean Licensor and any individual or Legal Entity
              on behalf of whom a Contribution has been received by Licensor and
              subsequently incorporated within the Work.
        
           2. Grant of Copyright License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              copyright license to reproduce, prepare Derivative Works of,
              publicly display, publicly perform, sublicense, and distribute the
              Work and such Derivative Works in Source or Object form.
        
           3. Grant of Patent License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              (except as stated in this section) patent license to make, have made,
              use, offer to sell, sell, import, and otherwise transfer the Work,
              where such license applies only to those patent claims licensable
              by such Contributor that are necessarily infringed by their
              Contribution(s) alone or by combination of their Contribution(s)
              with the Work to which such Contribution(s) was submitted. If You
              institute patent litigation against any entity (including a
              cross-claim or counterclaim in a lawsuit) alleging that the Work
              or a Contribution incorporated within the Work constitutes direct
              or contributory patent infringement, then any patent licenses
              granted to You under this License for that Work shall terminate
              as of the date such litigation is filed.
        
           4. Redistribution. You may reproduce and distribute copies of the
              Work or Derivative Works thereof in any medium, with or without
              modifications, and in Source or Object form, provided that You
              meet the following conditions:
        
              (a) You must give any other recipients of the Work or
                  Derivative Works a copy of this License; and
        
              (b) You must cause any modified files to carry prominent notices
                  stating that You changed the files; and
        
              (c) You must retain, in the Source form of any Derivative Works
                  that You distribute, all copyright, patent, trademark, and
                  attribution notices from the Source form of the Work,
                  excluding those notices that do not pertain to any part of
                  the Derivative Works; and
        
              (d) If the Work includes a "NOTICE" text file as part of its
                  distribution, then any Derivative Works that You distribute must
                  include a readable copy of the attribution notices contained
                  within such NOTICE file, excluding those notices that do not
                  pertain to any part of the Derivative Works, in at least one
                  of the following places: within a NOTICE text file distributed
                  as part of the Derivative Works; within the Source form or
                  documentation, if provided along with the Derivative Works; or,
                  within a display generated by the Derivative Works, if and
                  wherever such third-party notices normally appear. The contents
                  of the NOTICE file are for informational purposes only and
                  do not modify the License. You may add Your own attribution
                  notices within Derivative Works that You distribute, alongside
                  or as an addendum to the NOTICE text from the Work, provided
                  that such additional attribution notices cannot be construed
                  as modifying the License.
        
              You may add Your own copyright statement to Your modifications and
              may provide additional or different license terms and conditions
              for use, reproduction, or distribution of Your modifications, or
              for any such Derivative Works as a whole, provided Your use,
              reproduction, and distribution of the Work otherwise complies with
              the conditions stated in this License.
        
           5. Submission of Contributions. Unless You explicitly state otherwise,
              any Contribution intentionally submitted for inclusion in the Work
              by You to the Licensor shall be under the terms and conditions of
              this License, without any additional terms or conditions.
              Notwithstanding the above, nothing herein shall supersede or modify
              the terms of any separate license agreement you may have executed
              with Licensor regarding such Contributions.
        
           6. Trademarks. This License does not grant permission to use the trade
              names, trademarks, service marks, or product names of the Licensor,
              except as required for reasonable and customary use in describing the
              origin of the Work and reproducing the content of the NOTICE file.
        
           7. Disclaimer of Warranty. Unless required by applicable law or
              agreed to in writing, Licensor provides the Work (and each
              Contributor provides its Contributions) on an "AS IS" BASIS,
              WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
              implied, including, without limitation, any warranties or conditions
              of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
              PARTICULAR PURPOSE. You are solely responsible for determining the
              appropriateness of using or redistributing the Work and assume any
              risks associated with Your exercise of permissions under this License.
        
           8. Limitation of Liability. In no event and under no legal theory,
              whether in tort (including negligence), contract, or otherwise,
              unless required by applicable law (such as deliberate and grossly
              negligent acts) or agreed to in writing, shall any Contributor be
              liable to You for damages, including any direct, indirect, special,
              incidental, or consequential damages of any character arising as a
              result of this License or out of the use or inability to use the
              Work (including but not limited to damages for loss of goodwill,
              work stoppage, computer failure or malfunction, or any and all
              other commercial damages or losses), even if such Contributor
              has been advised of the possibility of such damages.
        
           9. Accepting Warranty or Additional Liability. While redistributing
              the Work or Derivative Works thereof, You may choose to offer,
              and charge a fee for, acceptance of support, warranty, indemnity,
              or other liability obligations and/or rights consistent with this
              License. However, in accepting such obligations, You may act only
              on Your own behalf and on Your sole responsibility, not on behalf
              of any other Contributor, and only if You agree to indemnify,
              defend, and hold each Contributor harmless for any liability
              incurred by, or claims asserted against, such Contributor by reason
              of your accepting any such warranty or additional liability.
        
           END OF TERMS AND CONDITIONS
        
           APPENDIX: How to apply the Apache License to your work.
        
              To apply the Apache License to your work, attach the following
              boilerplate notice, with the fields enclosed by brackets "[]"
              replaced with your own identifying information. (Don't include
              the brackets!)  The text should be enclosed in the appropriate
              comment syntax for the file format. We also recommend that a
              file or class name and description of purpose be included on the
              same "printed page" as the copyright notice for easier
              identification within third-party archives.
        
           Copyright 2026 Nikita Korvyakov
        
           Licensed under the Apache License, Version 2.0 (the "License");
           you may not use this file except in compliance with the License.
           You may obtain a copy of the License at
        
               http://www.apache.org/licenses/LICENSE-2.0
        
           Unless required by applicable law or agreed to in writing, software
           distributed under the License is distributed on an "AS IS" BASIS,
           WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
           See the License for the specific language governing permissions and
           limitations under the License.
        
Project-URL: Homepage, https://github.com/nkorvyakov28-AS/adaptersentry-m1
Project-URL: Documentation, https://github.com/nkorvyakov28-AS/adaptersentry-m1/tree/main/docs
Project-URL: Repository, https://github.com/nkorvyakov28-AS/adaptersentry-m1
Project-URL: Bug Tracker, https://github.com/nkorvyakov28-AS/adaptersentry-m1/issues
Project-URL: Changelog, https://github.com/nkorvyakov28-AS/adaptersentry-m1/blob/main/CHANGELOG.md
Keywords: security,llm,lora,adapter,safetensors,malware-detection,static-analysis
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Developers
Classifier: Intended Audience :: Science/Research
Classifier: License :: OSI Approved :: Apache Software License
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Programming Language :: Python :: 3.14
Classifier: Topic :: Security
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: Typing :: Typed
Requires-Python: >=3.11
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: safetensors<1,>=0.4.0
Requires-Dist: numpy<3,>=2.0
Requires-Dist: scipy<2,>=1.13
Requires-Dist: scikit-learn<2,>=1.5
Requires-Dist: pydantic<3,>=2.5.0
Requires-Dist: rich<16,>=13.0.0
Provides-Extra: bench
Requires-Dist: huggingface_hub<2,>=0.20.0; extra == "bench"
Requires-Dist: psutil<8,>=5.9.0; extra == "bench"
Provides-Extra: dev
Requires-Dist: adaptersentry[bench]; extra == "dev"
Requires-Dist: pytest<10,>=7.4.0; extra == "dev"
Provides-Extra: ray
Requires-Dist: ray[default]<3,>=2.9.0; extra == "ray"
Dynamic: license-file

<div align="center">

# 🛡️ AdapterSentry

**Static security scanner for LoRA adapters — catch a backdoor before the first trigger.**

[![Python](https://img.shields.io/badge/python-%3E%3D3.11-blue)](https://www.python.org/)
[![License](https://img.shields.io/badge/license-Apache%202.0-blue)](LICENSE)
[![Version](https://img.shields.io/badge/version-2.0.2-blue)](CHANGELOG.md)
[![Schema](https://img.shields.io/badge/ScanResult-2.1.0-6f42c1)](docs/output-schema/scan-result.md)

</div>

A LoRA adapter is a small file that changes how a large language model behaves. Anyone can
publish one, and a poisoned adapter can hide a **backdoor** that stays silent until a trigger
word appears in the input. AdapterSentry inspects the adapter's weights **before** it is
loaded — **without running the model** and **without knowing the trigger**.

- 🚫 **Runs nothing** — reads the `.safetensors` file only; never loads the model or executes code
- 🔍 **Needs no trigger** — works from the geometry of the trained update, not its behaviour
- 📦 **Scales to 70B–405B and MoE** — streamed in ~160 MB of RAM, on CPU
- 🔌 **CI-ready** — JSON and SARIF output, exit codes, a `--fail-on` gate

---

## How it works

```mermaid
flowchart LR
    F["📄 .safetensors<br/>+ adapter_config.json"] --> V["✅ Validate<br/>header, dtype,<br/>shape, bounds"]
    V --> I["🗂️ Inventory<br/>classify every<br/>tensor"]
    I --> X["🔬 Exact ΔW features<br/>per module, streamed<br/>via r×r cores"]
    X --> C["⚖️ Compare in family<br/>robust z vs its<br/>own siblings"]
    C --> S["🚩 Structural<br/>findings"]
    S --> D{Verdict}
    D -->|clean| A["🟢 allow"]
    D -->|anomaly / flag| R["🟡 review"]
    D -->|strict + red flag| B["🔴 block"]
```

| Stage | What happens |
|---|---|
| **① Validate** | The file must be a plain `.safetensors`; header length, dtype, shapes, byte ranges, element counts and LoRA rank are checked **before any tensor is allocated**. |
| **② Inventory** | Every tensor is classified — LoRA A/B pairs (layer, expert, module), full `modules_to_save` weights, DoRA vectors, trainable-token deltas, and anything unanalysable. Nothing is skipped silently. |
| **③ Exact features** | For ΔW = s·B·A (scale `s = α/r` from the config), singular values, norms and row/column norms are computed **exactly** through the tiny r×r matrices AAᵀ and BᵀB. ΔW itself — 235 M values for a 70B MLP layer — is **never built**. Features describe the *shape* of the update (stable rank, participation ratio, spectral entropy, entry kurtosis, row/column concentration), not its size. |
| **④ Compare** | Each module is compared with its family (all `q_proj`, all `down_proj`, …). The verdict looks for one pattern: an update **concentrated on a few outputs**, unlike its siblings (*concentration z*). Needs no reference data; insensitive to adapter strength. |
| **⑤ Structural** | Red flags numbers can't show: full `lm_head`/embedding/router matrices, LoRA on a MoE router, rank ≠ declared `r` (merged adapters), code/pickle/archive files next to the adapter. |
| **⑥ Verdict** | One fail-closed rule set → `allow` / `review` / `block` with reasons, confidence and limits. |

> 💡 **Why shape, not size?** A backdoor that fires on one rare trigger tends to pile the
> update onto a few specific outputs — like one instrument suddenly playing far louder than the
> orchestra. That concentration stays visible in the weights even when the trigger is unknown.

---

## Quick start

```bash
pip install adaptersentry

adaptersentry scan adapter_model.safetensors                      # text summary
adaptersentry scan adapter_model.safetensors --format json        # ScanResult 2.1.0
adaptersentry scan adapter_model.safetensors --format sarif \
    --output adaptersentry.sarif --fail-on review                 # GitHub code scanning + CI gate
adaptersentry batch --input-dir adapters/ --workers 8 --fail-on review
```

```python
from pathlib import Path
from adaptersentry import scan

result = scan(Path("adapter_model.safetensors"))
print(result.verdict.action, result.verdict.level)
```

> Keep `adapter_config.json` next to the `.safetensors` file (as PEFT saves it): the scan
> needs `lora_alpha` to know the real size of the update.

<details>
<summary><b>CLI options</b></summary>

| Option | Meaning |
|---|---|
| `--format text\|json\|full-json\|sarif` | output; `full-json` adds per-module features |
| `--policy default\|strict` | `strict` blocks on structural red flags |
| `--fail-on review\|block` | exit code `2` if the verdict reaches this action |
| `--mode full\|fast` | `fast` uses fewer samples for the kurtosis estimate |
| `--include-modules`, `--include-heads` | per-module / per-head detail in JSON |
| `--full-paths` | report the absolute path instead of the file name |

Exit codes: `0` completed · `1` adapter could not be analysed (verdict still `review`) ·
`2` `--fail-on` threshold reached.
</details>

---

## What you get

```text
$ adaptersentry scan adapter_model.safetensors
AdapterSentry 2.0.2 · adapter_model.safetensors
================================================================
Verdict:  REVIEW (MEDIUM) · confidence medium · behavioural check recommended
Status:   ok · policy default · mode full
Why:
  - [LOW] NO_REFERENCE_PROFILE: No calibrated reference for this model family
  - [MEDIUM] INTRA_OUTLIER: Update concentrated on a few outputs, unlike the rest of its
             family (down_proj layer 47): concentration z = 21.3 ≥ 14.0
Adapter:  peft_lora, family llama (config), 80 layers, r=64, α=128, scaling alpha_over_r
Coverage: 560/560 modules analysed; 0 tensor(s) not analysed
Intra:    concentration z 21.3 · max single-feature z 64.0 · 9 outlier module(s)
```
<sub>Illustrative output.</sub>

| Verdict | Meaning |
|:--:|---|
| 🟢 `allow` | No rule fired. **Not** a proof of safety. |
| 🟡 `review` | An intra-adapter anomaly, a structural flag, or part of the file was not parsed — look closer; behavioural check recommended. |
| 🔴 `block` | Structural red flag under `--policy strict`. The default policy never blocks. |

CI gates read `verdict.action`. Full field reference:
[ScanResult 2.1.0](docs/output-schema/scan-result.md) ·
[JSON Schema](docs/output-schema/scan-result-2.1.0.schema.json).

---

## Limits — read before relying on a verdict

- **A first filter, not a proof.** Weight-space detection is strong in the lab but bounded:
  detectors tuned on known attacks fail on new ones, and on large models the spread between
  honest adapters (different seeds) can exceed a backdoor's signal. A low-risk verdict is
  **not** evidence an adapter is safe — behavioural verification in isolation is still needed
  for untrusted adapters.
- **Calibration is partial.** The threshold (concentration z ≥ 14) was set on **393 popular
  Hub adapters** (Oct 2026, mostly Qwen/Llama ≤ 14B), presumed benign: about **5 %** get
  `review`. Detection was measured only on a **toy payload** inserted into real adapters —
  **71 %** caught when strong, **≤ 3 %** when weak. Real backdoors are not toy payloads;
  labelled measurement and per-family reference profiles are the next steps. Until then
  confidence is at most `medium`.
- **An attacker who knows the detector** can spread a payload across many directions, or pad it
  with inert components. Static analysis alone cannot fully rule this out.
- **Not analysed:** full weight matrices (need the base model), trainable-token deltas; DoRA
  partially. Each is reported and lowers confidence or raises the verdict to `review`.
- **Pickle formats** (`.bin`, `.pt`) are never deserialised — use a serialisation scanner.

---

## Performance

Measured on an 8-CPU server, synthetic Llama-3.3-70B-shaped adapter (80 layers × 7 modules,
r = 64, 1.66 GB bf16): end-to-end in **~18–20 s**, peak memory **~160 MB**. The 1.x engine
needed ~3.3 GB just to load the same file — the exact r×r cores are what let it stay this small.

---

## Testing & feedback

Trying AdapterSentry on your own adapters, or found a **false positive**, a **bug**, or a
**vulnerability**? Test results and reports are very welcome:

- **Telegram:** [@niknkor](https://t.me/niknkor)
- **Email:** [n.korvyakov28@gmail.com](mailto:n.korvyakov28@gmail.com)

A scan report helps a lot — attach the JSON:

```bash
adaptersentry scan ./adapter_model.safetensors --format json --output report.json
```

For a malicious adapter found in the wild, or a security issue, see also [SECURITY.md](SECURITY.md).

---

## Development

```bash
git clone https://github.com/nkorvyakov28-AS/adaptersentry-m1
cd adaptersentry-m1
uv sync --frozen --extra dev        # or: pip install -e ".[dev]"
pytest tests/ -q
```

[Architecture](docs/architecture/m1-architecture.md) ·
[CLI](docs/cli/usage.md) ·
[Batch engine](docs/architecture/scan-engine.md) ·
[CHANGELOG](CHANGELOG.md)

<details>
<summary><b>Migrating from 1.x</b></summary>

| 1.x | 2.x |
|---|---|
| `analyze()` / `scan_to_result()` → `AdapterReport` | `scan()` → `ScanResult` |
| `verdict.recommended_action` | `verdict.action` |
| `overall_score` (0–100) | removed — `verdict.level` + `anomaly` (z-scores) |
| `status`, `parse_status`, `analysis_mode` | `status` + `coverage` |
| `--format summary-json` / `debug-json` | `--format json` / `full-json` |
| `--fail-on LOW…CRITICAL` | `--fail-on review\|block` |
| `--rank` | removed — read from tensors, checked against config |
| `adaptersentry-m1` command | `adaptersentry scan` |

1.x results are not converted — re-scan the adapters.
</details>

AdapterSentry M1 is the open static-analysis layer. Planned: **M2 Behavioral Sandbox** —
advanced behavioral analysis.

## Background

The feature design follows published work on weight-space backdoor detection in LoRA adapters —
*Detecting Backdoored LoRAs from Weights Alone* (arXiv:2602.15195),
*Token-Level Generalization in LoRA Adapter Backdoors* (arXiv:2605.30189) and
Z-PEFT (arXiv:2608.02271): shape over magnitude, per-module and per-head features, MLP
coverage, and one-class calibration instead of a classifier trained on known attacks.

## License

Apache 2.0 — see [LICENSE](LICENSE).
