# Palladium

> Palladium is Owned Intelligence for the legal profession: legal AI that never leaves the firm's control. It is designed to run inside a law firm, on the firm's premises or in the firm's own cloud, on a model the firm owns, with keys the firm holds.

Palladium Intelligence Inc. is an Ontario corporation based in Toronto, Canada. Contact: info@getpalladium.ai.

## What Palladium does

- A secure internal model runs on the firm's infrastructure, sees documents in full, and handles most drafting, review, summarisation and search without anything leaving the firm.
- When a question needs more, a lawyer can send a redacted request to a frontier model (Claude, GPT, Gemini or Grok), or to all four at once through the Adversarial Frontier Council.
- The Council does not average answers. It returns where the models agree as confidence and where they disagree as exceptions, with a Defensibility Score, so the lawyer reviews only what the models contest.
- Every edit from the assistant arrives as a tracked change for a lawyer to accept or reject, and each decision is written to the matter's audit trail, which stays with the firm.

## How client details are protected

- Before any request leaves the firm, an internal model replaces client names, counterparties, individuals, addresses, account numbers, dates of birth and other identifying details with consistent placeholders.
- A second, separate check reads the outgoing request as an adversary would and confirms it cannot be reconstructed from what remains.
- Matches the engine is unsure about wait in a review queue, and the matter cannot go to an outside model until a lawyer has confirmed, corrected or dismissed each one.
- The key that links placeholders to real values stays inside the firm. Answers are restored inside the perimeter.
- Most AI vendors offer zero data retention (ZDR). Palladium is built on Zero Intelligence Retention™ (ZIR™): nothing the firm's lawyers do, including their reasoning, is kept outside the firm.

## Who owns what the system learns

The firm owns three layers: the redaction map for every matter, the working record of how its lawyers review and decide, and the model itself. All three stay on the firm's infrastructure, and none of it trains a model outside the firm. The firm-owned model and the learning flywheel arrive in Phase 2, from Q3 2027.

## How Palladium differs from Harvey, Copilot and other cloud legal AI

Most legal AI runs on the vendor's infrastructure, and its protection is a contract and a certification. Palladium's protection is the architecture: there is no Palladium server holding a firm's matters, and what the system learns improves a model only that firm can use.

## When Palladium fits

Palladium is built for:

- Firms whose clients' terms forbid their data being used to train AI.
- Firms that must keep client data in their own country, or inside their own walls.
- Firms whose partners will not put client files into a cloud AI tool because of privilege.
- Firms that want what their lawyers teach the AI to become an asset of the firm, not the vendor.

Palladium is not yet the right choice for:

- A solo lawyer who wants a low-cost monthly subscription today.
- A firm content to run all of its AI in a vendor's cloud.

## Palladium and cloud legal AI, side by side

| | Palladium | Typical cloud legal AI |
|---|---|---|
| Where documents are processed | Inside the firm, on its premises or its own cloud | On the vendor's servers |
| Who holds the keys | The firm | The vendor |
| What protects client data | The architecture | A contract and a certification |
| What an outside model sees | Placeholders, only when a lawyer chooses | The documents |
| Who owns what the system learns | The firm | The vendor |
| If the firm leaves | The deployment, the data and the model stay with the firm | The firm's history stays with the vendor |

## Questions to ask any legal AI vendor

1. Where are our documents processed, and on whose servers?
2. Who holds the encryption keys?
3. Could a breach at your company, or a subpoena served on you, reach our client files?
4. What do you keep after a request, including our lawyers' corrections?
5. Does anything we do improve a model that other firms use?
6. Does an outside model see our client names, or placeholders?
7. If we stop working with you, what do we keep?

## Next step

See Palladium on your own documents. Write to info@getpalladium.ai, or use https://getpalladium.ai/#contact. We reply within one business day.

## For investors

The thesis: frontier AI is becoming a commodity. The asset that cannot be copied is how a firm's lawyers reason. Palladium lets a firm put frontier AI to work on that reasoning without giving it away, and turns every matter a lawyer reviews into an asset the firm owns.

Why now:

- Open-weight models now come close to frontier quality at a fraction of the cost. Legora's own benchmark of 18 September 2026 scored DeepSeek V4 Flash at 1.02 times average quality for 0.12 times the cost (https://legora.com/bar-sept).
- In ILTA's 2026 technology survey, law firms ranked generative AI second among their security challenges, the first year it appeared on that list (reported by LawSites, 14 September 2026).

The company: Palladium Intelligence Inc. is an Ontario corporation based in Toronto and an AXL Venture Studio company, with an anchor pilot at a global law firm. Geoff Osler, Founder and CEO, previously worked at Apple and Adobe and took ClearSign public on NASDAQ.

Investor enquiries: info@getpalladium.ai, subject "Investors".

## Pages

- [Home page, plain text](https://getpalladium.ai/index.md): the full website content in Markdown.
- [Company facts](https://getpalladium.ai/facts.json): stable facts in machine-readable form.
- [Home page](https://getpalladium.ai/): the human version.
- [Request access](https://getpalladium.ai/#contact): demonstrations on documents like the firm's own.
- [The Palladium one-pager](https://getpalladium.ai/palladium-one-pager.pdf): a one-page overview of Palladium, as a PDF, to share with partners.