Skip to content

Releases: draiagent/AI-IP-Shield

Release list

AI-IP-Shield v0.1.0

AI-IP-Shield v0.1.0 Latest
Latest

Choose a tag to compare

View all tags
@draiagent draiagent released this 19 Aug 06:14
v0.1.0
56fe0f6

AI-IP Shield v0.1.0 — Public Alpha

繁體中文 · English

繁體中文

首次公開 Alpha

AI-IP Shield v0.1.0 是第一個可公開測試的完整版本，整合圖片與 PDF 隱形追蹤、Fingerprint Registry、SHA-256 完整性、Ed25519 來源證明、Attack Lab、Evidence Fusion，以及 ChatGPT / Codex / Claude Code Agent Skill。

包含功能

  • JPG / PNG 指紋與隱形浮水印保護
  • PDF raster-visual 逐頁追蹤
  • SQLite Fingerprint Registry
  • SHA-256 exact integrity check
  • Ed25519 sidecar provenance + local trust store
  • optional C2PA adapter
  • Do Not Train policy assertion
  • 可重現 Attack Lab / release gates
  • Codex / Claude Code repo/user scoped Skill
  • ChatGPT portable Skill ZIP
  • CLI protect、verify、attack、provenance、Skill 管理命令

已驗證 baseline

固定 image engine 曾通過專案定義的 100-image / 500-negative-control release gate；PDF adapter 曾通過 20-PDF / 100-page gate。這些結果是專案內部工程 benchmark，不是第三方安全認證，也不代表所有真實攻擊都能達到 100%。

已知限制

  • PDF V0.1 採 raster-visual 模式，不保留原生文字／向量語意。
  • Native C2PA smoke test 仍依執行環境而定。
  • PixelSeal 仍是 challenger engine，待真實 checkpoint benchmark。
  • Camera capture、Perspective、AI editing / regeneration 與 anti-distillation 為後續測試範圍。

English

First public alpha

AI-IP Shield v0.1.0 is the first public testing release combining image/PDF invisible traceability, a fingerprint registry, SHA-256 integrity checks, Ed25519 provenance, Attack Lab, evidence fusion, and portable Agent Skills for ChatGPT / Codex / Claude Code.

Included

  • JPG / PNG fingerprint and invisible watermark protection
  • PDF raster-visual page-level traceability
  • SQLite fingerprint registry
  • SHA-256 exact integrity checks
  • Ed25519 sidecar provenance and local trust store
  • optional C2PA adapter
  • Do Not Train policy assertion support
  • reproducible Attack Lab and release gates
  • repo/user-scoped Codex and Claude Code Skills
  • portable ChatGPT Skill ZIP
  • CLI for protect, verify, attack, provenance, and Skill management

Validated baseline

The pinned image engine previously passed the project's defined 100-image / 500-negative-control release gate. The PDF adapter previously passed the defined 20-PDF / 100-page gate. These results are internal engineering benchmarks, not third-party security certifications and not guarantees of 100% robustness against every real-world attack.

Known limitations

  • PDF v0.1 is raster-visual and does not preserve native text/vector semantics.
  • Native C2PA smoke testing remains environment-dependent.
  • PixelSeal remains a challenger engine pending a real checkpoint benchmark.
  • Camera capture, perspective distortion, AI editing/regeneration, and anti-distillation remain future test areas.

完整原始碼現已同步至 main branch：https://github.com/draiagent/AI-IP-Shield

Assets 7
Loading