shinrai-pii-m-v1.1

<img src="https://huggingface.co/spaces/innovius/README/resolve/main/shinrai-logo.png" width="104" alt="ShinrAI — shield logo" align="right">

shinrai-pii-m-v1.1

You use the AI. The AI does not use you.

Superseded by shinrai-pii-m-v1.3 and by ShinrAI 1.5 in the managed API at getshinrai.com for every language — including Japanese, where v1.3 reads its best result ever (98.1 register F1 vs 83.8 macro here). v1.1 remains available as the archived first open generation.

🔓 License: v1.1 is and remains Apache 2.0 — the permanently open tier. The v1.3 flagship ships under the Innovius Open License v1.0 (free for government, education, research, and companies under $10M revenue; it converts to Apache 2.0 on 2028-09-01).

M-size ShinrAI encryption model — the detection layer of ShinrAI semantic encryption (ShinrAI, 信頼, is Japanese for trust). This is the model generation behind our 2024–2026 ShinrAI Enterprise product line (legacy), and the first ShinrAI encryption model we opened up. German and English were first superseded by shinrai-pii-m-v1.2; all three of its languages now route to v1.3.

Architecture: ModernBERT-family mmBERT-base encoder with a multi-head token classifier — PERSON / CITY / STREET / ORG spans with frequency/size tiers and span attributes (origin, gender expression, name part) for realm-preserving pseudonymization: detected entities are swapped for natural alternatives matched to context, culture and language (never [MASK] wreckage, never bias-triggering hard cuts), so downstream AI keeps reasoning over natural text. Suited to mass de-identification (ePA / clinical, high-security, banking) as well as realtime protection.

Developed by Innovius with the EECC Research Labs, trained on the JURECA supercomputer at the Jülich Supercomputing Centre (Forschungszentrum Jülich) under the WestAI compute grant, scaling onto JUPITER, Europe's first exascale system.

• 🚀 ShinrAI 1.5 — the current model, in the managed ShinrAI PII detection and pseudonymisation API: getshinrai.com
• 🛡️ Try the current generation live (text, PDF, image, audio): shinrai-pii-playground
• 📊 Benchmarks: ShinrAI 1.5 vs Azure, Presidio, GLiNER & the open field (losses included): shinrai-pii-benchmarks
• 🧪 Beta programme: https://innovius.ai/shinrai/beta/
• 🏢 Organization: https://huggingface.co/innovius
• ✉️ Production use today (Secure AI Suite / ChtSafe): https://innovius.ai/#contact

What it does

Given text in German, English or Japanese, the model emits entity spans with rich attributes: entity class, a frequency/size tier (how common a name is, how large a city is), cultural origin, gender expression and name-part structure. This is what enables realm-preserving replacement: Robert → Paul — a substitute matching the original's cultural register, frequency tier and gender expression, with no visible trace that anything was replaced and minimal bias distortion for the consuming model.

Plus three span-pooled attribute heads computed in the same forward pass: origin (42 classes, independent of text language — "Mailand" in German text is a CITY with origin IT), gender expression (masc / fem / neutral / unknown; replacement never flips masc↔fem) and name part (given / family / full). The full machine-readable schema ships in labels-v2.0.yaml.

Where it sits — and what belongs in front of it

ShinrAI models exist for the multi-dimensional, context-dependent part of the problem: knowing that Keller is a surname here and a basement there, that Lindenweg 12 is an address, how identifying each entity is, and what a culturally consistent replacement looks like. That is what statistical models are for.

Deterministic identifiers are not. IBANs, credit-card and account numbers, BTC/ETH wallet addresses, postal codes, phone numbers, e-mail addresses, tax and insurance IDs follow fixed formats, most with checksums — catch and replace them in an application layer in front of the model (validators, regex, lookup tables). Hunting fixed patterns with a 300M-parameter encoder is highly error-prone compared to a validator that is exact by construction, and a waste of resources — RAM footprint and response time — on something a rule decides in microseconds; anything the validator removes can never leak past it. Our own software layer does exactly that: the ShinrAI Connector ships it on your desktop, and the Secure AI Suite ([email protected]) brings it to your company — this model only ever owns what rules cannot decide.

Results

Span F1 on surface-disjoint evaluation suites (never seen in training, disjoint surface forms), measured for this exact checkpoint:

Attribute accuracy on detected spans (these steer replacement selection):

Known limitations (addressed in the current generation)

• ORG generalization is the weak head (56.6–68.9): the organization inventory this generation trained on was small. v1.2 moved to a Wikidata-scale inventory and reads 88.4–93.8.
• Under-recall on institutional letterhead address blocks, abbreviated given names ("M. Krüger") and letter sign-off names.
• PERSON spans are emitted at given/family granularity (e.g. "Anna" + "Keller" as two spans); v1.2 merges full names into one span.
• German and Japanese PERSON tiers lean towards common; tier depth improves with the registry-backed statistics introduced in v1.2.
• Context window 512 tokens (sliding-window inference with overlap merge handles longer documents automatically; v1.2 doubles the window to 1024).

Routing guidance: all three languages → shinrai-pii-m-v1.3 (de 94.4 / en 97.7 / ja 98.1 register F1, plus 12 further locales and 15 format heads).

How it compares

Against public PII models. We benchmarked the strongest openly released PII models we could find on this model's serving suites — the OpenMed-PII family (clinical-domain de-identification, RoBERTa/DeBERTa-class encoders), taking the best in-language checkpoint per suite (SuperClinical-434M / BigMed-560M class). Identical scoring pipeline, span-level concept intersection over {PERSON, CITY, STREET, ORG}, measured 2026-08. Macro span F1:

Read it fairly, in both directions:

• Our suites are multilingual business/medical documents — outside OpenMed's clinical home domain. On English their flagship is genuinely competitive on PERSON/CITY/STREET (81–89 F1); the macro gap there is mostly ORG, a systematic blind spot of every PII tagger we measured (16–39 F1 across the three suites).
• On their home turf — real German clinical narrative (GraSCCo, which is eval-only for us) — OpenMed's German flagship beats this model on PERSON recall and aggregate F1. If your workload is pure clinical de-identification inside OpenMed's domain, a clinical specialist is a strong choice there; our Enterprise ePA pipeline layers this model with additional safeguards for exactly that reason.
• Generic RoBERTa/DeBERTa-class PII taggers solve a different, smaller problem: flat entity labels, usually English-first. None of the models we measured emit what replacement needs — frequency/size tiers, cultural origin, gender expression, name-part structure, calibrated confidences. Detection is table stakes; ShinrAI's job is making the replacement invisible.

Against the next ShinrAI generations. shinrai-pii-m-v1.2 superseded this model for German and English — 92.2 / 94.3 macro on suites drawn from far larger entity inventories, where this model reads 33.6–49.3 — and added Italian, French, Spanish and Polish (Japanese stayed here at the time: 83.8 vs v1.2's 60.8). shinrai-pii-m-v1.3 now supersedes both everywhere, Japanese included (98.1 register F1), with 15 locales and 15 format heads on top of the four semantic heads.

Bias behavior — engineered for model-bias stability

This model family is trained to avoid machine biases — the documented, systematic biases of LLMs — not to correct human bias in text. Large models measurably treat text differently depending on name origin, gender association, name rarity, city size or organization prestige. Naive anonymization changes those signals: [MASK] tokens, <PII> tags and randomly chosen substitute names all shift how the consuming model scores, ranks and reasons over a document — which makes results unstable and the pseudonymization itself detectable.

The replacement policy is therefore bias-stable by construction:

• An entity is never swapped for one that triggers more or fewer known model biases than the original. Replacements match the original's cultural origin, gender expression, frequency tier and register — same treatment by the downstream model, different identity.
• No debiasing. If the original text would trigger a model bias, the replacement triggers the same one. Silently "improving" the text would change model behavior and break the round-trip guarantee.
• The tier and attribute heads exist precisely for this: they are the steering signal that keeps the statistical texture of text constant under replacement — never a profiling signal (see Privacy below).

Quantization — run it anywhere

The full ONNX ladder ships in quant/ (opset 17; export parity vs PyTorch: max logit delta 3.5e-05 across all 7 output heads — quant/parity-onnx.json):

Q8 is the serving baseline: ≤ 0.2 macro-F1 cost on every serving suite, 4× smaller, CPU-friendly. Q4 costs ~2 macro points where measured — usable where 257 MB matters.

Usage

The repo is a self-contained HF-layout release:

model.safetensors      # full-precision weights (custom multi-head architecture)
config.json            # encoder configuration (ModernBERT / mmBERT-base)
tokenizer/             # HF tokenizer (256k vocab)
labels-v2.0.yaml       # machine-readable label schema (heads, tiers, attributes)
calibration.json       # per-head confidence temperatures
quant/                 # ONNX ladder + MANIFEST.json + parity report
assets/                # model card figures

The heads are a custom multi-head architecture, so plain AutoModelForTokenClassification does not expose them — use the ONNX graph (named inputs/outputs, works with onnxruntime alone), or the ShinrAI inference SDK (sliding-window batching, IOB2 decoding, span merging, calibration — pip package planned):

import onnxruntime as ort
from transformers import AutoTokenizer

repo = "innovius/shinrai-pii-m-v1.1"  # local: path to a clone
tok = AutoTokenizer.from_pretrained(repo, subfolder="tokenizer")
sess = ort.InferenceSession("quant/model-q8.onnx")

enc = tok("山田花子さんは大阪市に住んでいます。", return_tensors="np")
out = sess.run(None, {"input_ids": enc.input_ids, "attention_mask": enc.attention_mask})
# 7 named outputs:
#   logits_person / logits_city / logits_street / logits_org   [batch, seq, 7]  (IOB2 × 3 tiers + O)
#   attr_logits_origin [.., 42] · attr_logits_gender_expression [.., 4] · attr_logits_name_part [.., 3]
# label inventories for decoding: labels-v2.0.yaml

Decoded output (via the ShinrAI predictor, verified against this exact checkpoint) for "山田花子さんは大阪市に住んでいます。":

{"span": [0, 4], "text": "山田花子", "type": "PERSON", "tier": "common",
 "confidence": 0.753, "attrs": {"origin": "JP", "gender_expression": "unknown", "name_part": "family"}}
{"span": [7, 10], "text": "大阪市", "type": "CITY", "tier": "major",
 "confidence": 0.993, "attrs": {"origin": "JP"}}

and for "Bitte senden Sie den Bericht an Frau Dr. Anna Keller, Lindenweg 12, 04109 Leipzig.":

{"span": [41, 45], "text": "Anna",         "type": "PERSON", "tier": "common", "confidence": 0.781}
{"span": [46, 52], "text": "Keller",       "type": "PERSON", "tier": "common", "confidence": 0.947}
{"span": [54, 66], "text": "Lindenweg 12", "type": "STREET", "tier": "generic", "confidence": 0.668}
{"span": [74, 81], "text": "Leipzig",      "type": "CITY",   "tier": "medium",  "confidence": 0.995}

Note what is not in the output: the postal code 04109. Deterministic identifiers are your validation layer's job (see the pipeline section) — by design, not omission.

Inference is sliding-window (window 512, stride 64, overlap merge) — documents of any length work. Confidences are temperature-calibrated per head from calibration.json.

Use ShinrAI in production today

☁️ As a developer or company — the ShinrAI API. ShinrAI 1.5, the current model, runs in the managed ShinrAI PII detection and pseudonymisation API, hosted on STACKIT in the EU. Start at getshinrai.com — see the ShinrAI plans and pricing and the API documentation.

🖥️ On your own machine — ShinrAI Connector. Our desktop client runs ShinrAI models locally on your own GPU — or CPU-only, thanks to the Q8 build — and funnels your AI traffic through semantic encryption before it leaves your machine. Works with Claude Desktop and Cursor today, the ChatGPT Desktop app soon. Get it at chtsafe.com/apps.

The Connector running this exact model on a Mac — 0.12 GB in memory, CPU-only — with the deterministic categories (banking and cards, crypto wallets, ID numbers, contact details) handled by its software layer, exactly as described above.

🔒 As a private person — ChtSafe. Chat fully anonymously with any leading AI model, protected in realtime by the latest ShinrAI models (currently in review and training, publication to follow). Sign up at chtsafe.com or get the app from the App Store.

🏢 As a company — the Secure AI Suite. Semantic encryption plus onion routing across as many model-access paths as you like, deployed on-premise first — your hardware, your keys — or in your private cloud. You get the complete software and code for review; we support the rollout on your infrastructure, or deliver custom GPU servers with 5-year support. Register your company at chtsafe.com or write to [email protected].

Training data

The v1.1 corpus is mostly hand-compiled — built from our own curated name, place and organization inventories, document templates and formatting frames — and synthetically reviewed: cross-checked and quality-gated by openly released models (the Qwen and Gemma families), with substantial review passes on top. It is not a fully synthetic corpus; it is a curated one, anchored in open data: GeoNames (CC-BY) city statistics and US SSA/Census name statistics (public domain).

Everything in the pipeline is either self-contributed or license-clean open data — no customer data, no scraped personal records, nothing proprietary, nothing "found" on the web. Every record carries a license-provenance field; released datasets contain only the license-clean subset (CC0 / CC-BY / public-domain / our own text). Data releases: https://huggingface.co/innovius (as they land).

Privacy & responsible use

The model detects and helps pseudonymize personal data; it does not memorize real individuals — the training corpus is hand-compiled and synthetically reviewed, with no real persons' records in it. Detection is probabilistic: for regulated deployments, treat it as one layer of a defense-in-depth privacy architecture, not a sole control (and keep deterministic identifiers in your validation layer, per above). Attribute predictions (origin, gender expression) exist to preserve the statistical texture of text under replacement — never use them to profile real people.

Credits

EECC Research Labs · Forschungszentrum Jülich (JSC, JURECA & JUPITER) · WestAI · mmBERT (JHU CLSP, MIT) · the ModernBERT lineage · Qwen · Gemma · GeoNames · the US SSA and Census Bureau open statistics.

Citation

@software{shinrai_pii_2026,
  author  = {Innovius AI and EECC Research Labs},
  title   = {ShinrAI PII Models: tier- and bias-aware PII detection for realm-preserving pseudonymization},
  year    = {2026},
  version = {1.1},
  url     = {https://huggingface.co/innovius}
}