model-theft

model-theft is a cursor rule for Cursor from davidmatousek/tachi. It costs 43 tokens per session (2,270 once invoked), scanned A, original, Apache-2.0.

An AI security rule for detecting attempts to steal or recreate a language model and its private files, such as model weights or checkpoints.

In plain words
What is it for?
Use it to review model storage, inference APIs, registries, training systems, and fine-tuned models for theft risks.
Why use it?
It helps identify direct file theft, copying a model through repeated API queries, exposed model artifacts, and attacks that reveal model structure or parameters.

Cursor rule for Cursor

Written for Cursor: a Cursor rule (.mdc).

Needs its repository: it reads a path above its own folder, which exists only inside the repository. The line is output_schema: ../../../schemas/finding.yaml.

Good fit Use it to review model storage, inference APIs, registries, training systems, and…

Compare 6 cursor rules from other repositories ↓
Install

Getting it into your agent

It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.

Clone the repo
git clone --depth 1 https://github.com/davidmatousek/tachi
agentmods
npx agentmods add rules/davidmatousek/tachi/model-theft

Made for: Cursor.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for model-theft

README.md
[![agentmods](https://agentmods.dev/badge/rules/davidmatousek/tachi/model-theft.svg)](https://agentmods.dev/rules/davidmatousek/tachi/model-theft)
Your own site
<a href="https://agentmods.dev/rules/davidmatousek/tachi/model-theft"><img src="https://agentmods.dev/badge/rules/davidmatousek/tachi/model-theft.svg" alt="Measured on agentmods" height="20"></a>
Per session 43 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 2,270 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00043 $0.02270
Opus 5 $0.00022 $0.01135
Sonnet 5 $0.00009 $0.00454
Haiku 4.5 $0.00004 $0.00227

Measured 3d ago against content hash a207b7d00b66, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade A, and why

model-theft scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

adapters/cursor/rules/model-theft.mdc · 185 lines

How it starts

The opening of the file, as written. The whole thing — 185 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Metadata

category: llm
threat_class: LLM
dfd_targets: [Data Store, Process]
owasp_references: [OWASP LLM06:2026, OWASP LLM04:2026]
output_schema: ../../../schemas/finding.yaml

Model Theft Threat Agent

Purpose

Detects threats where an attacker attempts to steal, replicate, or extract proprietary model assets. Model theft encompasses direct exfiltration of model weights and parameters, API-based model extraction where systematic querying reconstructs a functional copy, and unauthorized access to model artifacts stored in training infrastructure. Successful model theft results in loss of intellectual property, enables adversaries to discover model vulnerabilities offline, and eliminates competitive advantages derived from proprietary model capabilities. This agent identifies weight exfiltration, API-based extraction, model artifact exposure, and side-channel attacks that reveal model architecture or parameters.

Detection Scope

Trigger Keywords

This agent activates when a DFD element name or description matches any of the following patterns (case-insensitive):

  • LLM
  • model
  • GPT
  • Claude
  • weights
  • checkpoint
  • inference
  • model registry
  • model serving
  • model API
  • fine-tuned

Applicable DFD Element Types

  • Data Store: Model registries, weight storage systems, checkpoint repositories, artifact stores, and any storage containing model parameters, configurations, or training outputs.
  • Process: Model serving endpoints, inference APIs, training pipelines, and fine-tuning processes that have access to model weights or produce outputs from which model behavior can be inferred.

Detection Patterns

  1. Direct Weight Exfiltration: Unauthorized access to stored model files, parameters, or checkpoints. Look for:
    • Model weight files stored in shared storage without access controls (S3 buckets, NFS mounts, model registries)
    • Overly broad IAM permissions on model artifact storage
    • Training pipelines that write checkpoints to world-readable locations
    • Model serving infrastructure where the container filesystem exposes weight files
    • Absence of encryption at rest for model artifacts

Read the full file on GitHub · 185 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 3d ago First seen · 185 lines · 43 tokens per session scan A a207b7d00b66

Subscribe to this mod's changes

model-theft is a cursor rule published in the GitHub repository davidmatousek/tachi (90 stars, last pushed 25d ago), licensed Apache-2.0. It adds 43 tokens to every session and 2,270 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.