# A control you can show an auditor, not a line in a prompt

Lock a tile. Prove you locked it. A rule in a system prompt is a request; Tileward Governance is a gate in front of the model that decides before the model writes anything and records every decision. This page covers the live demo, what a lock does, how it was tested, and the full list of lockable tiles.

Page: https://tileward.com/governance/

## Headline results

- [98% stopped](https://tileward.com/evals/#policy-bypass) of 2,251 requests written to get past a lock.
- [~14 tokens](https://tileward.com/research/governance-security-control/) for the gate to answer, before the model writes one.
- 215 lockable tiles, 64 of them regulated topics rather than harm categories.

## What a lock does

- **Refused before inference:** a request on a locked tile never reaches the model, so the completion is not billed and nothing off-policy is generated.
- **Matches the tile, not a word list:** a differently worded request in the tile's own vocabulary lands on the same lock; the same question in plain words with none of that vocabulary can get through.
- **Bound to the API key:** policy travels with the credential, not the prompt. One base-URL change and every call is governed.
- **Auditable:** a record is written for every decision, with the time, the decision, the tiles it matched, how the match was made, the size of the request, the API key or access token that made it, the actor your client names in its request header, and, where they apply, the policy version in force, the invoking model and the target model. CSV and JSON on every account; webhook and SIEM on Enterprise.

The page has a live demo against the same guard engine the API runs, with no account needed: lock a tile, type a question, and read the verdict, the tile a refusal matched and the tokens the guard read to decide. A calculator estimates what stops being billed when requests on a locked tile never reach the model.

## How it was tested

- **Requests written to get past a rule:** 2,201 of 2,251 were stopped before the model saw them: the same locked request reworded, reframed, split across sentences, or buried beside ordinary work. 50 crossed; most were a locked request riding inside an ordinary allowed one.
- **Ordinary work let through:** 45 of 50 benign controls were allowed on the same corpus; 10% were blocked.

The [write-up](https://tileward.com/research/governance-security-control/) and the [eval](https://tileward.com/evals/#policy-bypass) carry the method.

## 64 regulated topics, not 14 harm categories

Content guards answer "is this harmful?"; a regulated business needs "is this about a topic we are not licensed to discuss?" Harm guards are organized around 3–4 categories (ShieldGemma, Azure Content Safety) up to 23 (NemoGuard ContentSafety), with Llama Guard 4 at 14, where finance, medicine and law collapse into one bucket, S6 "Specialized Advice". Tileward ships AML & KYC, options pricing, personal medical advice and 61 more regulated topics, each lockable on its own. A harm filter and a Governance policy are not substitutes: Llama Guard is free, and you should run it too.

## The roster: 215 lockable tiles

64 regulated topics (the ones a licence, a regulator or a compliance officer turns on) and 151 capability tiles (programming languages, frameworks, the sciences, reasoning, general knowledge), each independently switchable per API key, with the same lock and the same audit record. Any account can add tiles of its own. The 64 regulated topics:

- **Finance (18):** AML & KYC, Client onboarding, Commodity trading, Conduct & surveillance, Corporate lending, Credit risk, Execution & trading, FX trading, Market risk, Operational risk, Options pricing, Personalized investment advice, Private equity & VC, Quant strategies, Regulatory reporting, Reporting & analytics, Structured products, Treasury & cash management
- **Safety (8):** Biosecurity & bioweapons, Child safety, Fraud & scams, Hate & harassment, Self-harm & crisis, Sexual & adult content, Surveillance & stalking, Weapons & dangerous goods
- **Industries (8):** Education, Energy & utilities, Entertainment & media, Manufacturing & supply chain, Pharma & life sciences, Real estate, Telecom & networks, Transportation & logistics
- **Legal (6):** Data privacy & PII, Employment law, Intellectual property, Legal & contracts, Litigation & disputes, Personal legal advice
- **Healthcare (5):** Clinical medicine, Health Insurance & Claims, Medical Devices, Personal medical advice, Public Health & Epidemiology
- **Business functions (5):** Customer support, IT & data operations, Marketing, Product management, Sales
- **Tax & insurance (3):** Insurance underwriting, Personal tax advice, Tax accounting
- **Civic & world (3):** Geopolitics & diplomacy, News & current events, Political affairs
- **Content integrity (3):** Copyright & piracy, Deepfakes & synthetic media, Misinformation & disinfo
- **Security (2):** Cybersecurity defense, Exploits & malware
- **Public sector & HR (2):** HR & employment, Public procurement
- **Technology (1):** Software engineering

## Pricing

Governance is included in every Tileward account. You pay for model usage at each model's rate, starting with a $25 credit and no card. [See the rates](https://tileward.com/pricing/), or [start free](https://app.tileward.com/?join=explore).

## Related

- [Governance write-up](https://tileward.com/research/governance-security-control/): how a gate before inference turns policy into a testable control
- [Policy-bypass eval](https://tileward.com/evals/#policy-bypass): the measured result and its method
- [Security](https://tileward.com/security/): what is kept, and where
- [Connect guide](https://tileward.com/docs/#connect): a recipe for each supported client
- [Talk to an engineer](https://calendar.app.google/hg2DzFTxKJgwFWm19): bring something you can't let the model discuss; we lock it, try to get past it, and show you the record