Is it safe to let an AI agent send contracts?
Published
Short answer
It is safe when the tool is built so the agent can prepare everything but a person decides the important steps. Look for five things: the agent acts with your own permissions, you can limit it to read-only, it can't delete anything, it never sends a contract without your say-so, and every change is logged.
What could go wrong
It helps to be concrete. With an agent connected to your contracts, the real risks are:
- Sending the wrong thing. A draft with a wrong price or the wrong client goes out.
- Doing too much. The agent changes or removes contracts you didn't mean to touch.
- Seeing too much. It reads contracts from a workspace or client it shouldn't.
- No trail. Something changed and nobody can tell who or what did it.
Each of these has a simple safeguard.
Five safeguards to look for
1. The agent acts as you
When you connect, you sign in with your own account. The agent can only do what you could do yourself, in the workspace you picked. If you can't send contracts in your role, neither can your agent.
2. You choose read-only or read and write
Some uses only need reading: "which contracts end this quarter?" A read-only connection can search and read but can't change anything.
3. Nothing can be deleted
The safest delete button is the one that doesn't exist. Pledge offers its agents no delete tools at all.
4. Sending needs you
Sending a contract is the step that matters: it locks the contract and emails the signers. In Pledge, it only happens when you press Send in the chat or tell the agent to send it. The agent can draft and edit as much as it likes before then.
5. Every change is logged
Changes made through an agent are recorded on the contract as made through an AI app, next to the changes people made by hand.
Questions to ask any tool
| Question | A good answer |
|---|---|
| Whose permissions does the agent use? | The signed-in person's, never an admin key |
| Can I limit it to reading? | Yes, at connect time |
| Can it delete data? | No, or only with confirmation |
| Can it send or sign without me? | No |
| Is there a log of what it did? | Yes, on each record |
| Does the tool get my conversations? | No, only the actions the agent takes |
| Can I disconnect it? | Yes, any time |
And the drafts themselves?
Safety isn't only about permissions. An agent drafting from a blank page can invent terms. An agent drafting from your approved templates mostly fills in fields. That's the difference between a draft you need to rewrite and one you only need to read. See why your agent should draft from your templates.
Questions
- Can an AI agent send a contract without asking me?
- It depends on the tool. In Pledge it can't: sending only happens when you press Send or tell the agent to send.
- Can the agent delete my contracts?
- Not in Pledge. There are no delete tools available to agents.
- Does Pledge see my chats with Claude or ChatGPT?
- No. Pledge only receives the actions the assistant takes, such as creating or sending a contract, not your conversation.
- How do I disconnect an agent?
- In Pledge, open Settings, then Agents, and disconnect the app. You can also remove the connector in the assistant itself.
Try Pledge
Contracts, time, plans and leads for small agencies, with an MCP server for Claude and ChatGPT. Pledge is in private beta: leave your email and we will send an invite.