What are ethical walls for AI?
Ethical walls for AI are barriers that stop an AI agent from using information across a line the company has drawn, such as between two client matters or between deal teams and research. The wall holds in every agent: for anyone on the wrong side, the walled information behaves as if it does not exist.
Last updated 25 October 2026 · Mert Uzun, co-founder
Why it matters
97%
of organizations that had an AI-related breach lacked proper AI access controls. (IBM, Cost of a Data Breach 2025)
$10.22M
The average cost of a data breach in the US in 2025. (IBM, Cost of a Data Breach 2025)
$670K
About that much more per breach for organizations with high levels of shadow AI. (IBM, Cost of a Data Breach 2025)
How it works in practice
Law firms, banks and advisers have always kept walls. A team on one side of a deal must not see the other side's work. Until now the wall lived in a memo and in folder settings.
AI agents cross folders easily. One question can pull from every matter the agent can reach, and the person asking may be on the wrong side of a wall.
An ethical wall for AI puts the rule into the answer itself. The general counsel sets the wall once, and every agent in the firm respects it.
For people inside the wall, nothing changes. For everyone else, the matter does not appear: not in answers, not in sources, not in lists.
How Mithrandir does it
Walls you set apply on top of every tool's own permissions and your hierarchy.
Every agent respects them: Claude, ChatGPT, Copilot, Cursor and 16 more.
Restricted content never reaches a model. 0 leaks across 30,000+ entries.
Questions
- Who sets an ethical wall?
- The people who own the rule today, usually the general counsel, compliance or a practice lead. They name the matter or team and who is inside.
- Can a wall cover part of a tool?
- Yes. A wall can cover one matter, one client, one channel or one folder, without changing anything else in that tool.
- What does a person outside the wall see?
- Nothing about it. The walled information does not appear in answers or sources, so the agent cannot hint that it exists.
- Is this only for law firms?
- No. Banks, advisers, fintech companies and any company with deal teams, investigations or confidential projects use the same idea.
