Attorney-Client Privilege
Your prompts can become discoverable evidence.
OpenAI has been ordered by a federal court to preserve and hand over millions of ChatGPT conversation logs, and separately complies with legal process for individual users' chat data. Legal AI wrappers route through those same servers. Most firms haven't asked this question yet. You should.
The Reality
The question most firms haven't asked
The legal AI tools your team is using today. What happens to that data when someone issues a subpoena?
“A federal court has ordered OpenAI to preserve and hand over millions of ChatGPT conversation logs.”
This happened. Ask your clients if they know.
OpenAI stores everything
ChatGPT retains chat history, uploaded PDFs, and documents by default, for up to two years. OpenAI complies with civil subpoenas and law-enforcement legal process for individual users' chat histories. This is not a hypothetical.
Wrappers route through the same servers
Legal AI wrappers are built on OpenAI and Anthropic APIs. When your data flows through those tools, it flows through those servers, the same servers that have been subpoenaed. The wrapper does not change the underlying risk.
ABA Model Rule 1.6 applies
Your professional responsibility obligations do not pause because you used AI. Model Rule 1.6 requires competent supervision of the technology you use for client work, including understanding whether your tools can maintain confidentiality.
The Wrapper Problem
The chain your data travels
Typical AI architecture
The wrapper company may have a DPA. But your data still flows through OpenAI or Anthropic infrastructure, the same infrastructure courts have ordered to preserve and produce user data.
Irys architecture
~80% of AI processing never leaves Irys-controlled infrastructure. When foundation models are used, only an encrypted ephemeral token is sent. Your client content is never transmitted or retained.
How Irys Is Different
Built to protect privilege from the architecture up
Privilege protection is not a feature we added. It is a constraint we built the entire system around.
Data processor, not data retention company
Irys processes your data to complete tasks, and retains it only under our data policies and the regulations that apply to us. There is no accumulated store of your client communications, documents, or strategies sitting on third-party servers.
~80% processed in-house
The majority of AI processing runs on Irys-controlled infrastructure. When foundation model calls are required, they use ephemeral encrypted tokens. No content is retained by the model provider after the session ends.
Never trained on
Your data is never used to train models, ours or anyone else's, and access is limited to your firm. Retention is governed by our data policies and the regulations that apply to us.
No training corpus
A subpoena reaches data that exists. Irys never pools your client data into a training corpus, and what we hold is retained only under our data policies and the regulations that apply to us.
See the full security architecture.
SOC 2 Type II, ISO 27001, no training on your data, and the technical details behind how Irys protects your clients.
