How Otto is governed.
The honest answer to “is this AI going to lie to my customers or do something I didn’t approve?” Here’s the line between what Otto does on its own and what always waits for a human.
Otto acts on its own
Low-risk, reversible, nothing a customer receives or that moves money.
A human must approve
Anything a customer sees, any money that moves, anything destructive.
High-risk outputs get human review
The more consequential the action — money, a customer’s inbox, anything hard to undo — the firmer the human gate. Otto can prepare it; a person releases it.
Otto discloses it’s an AI
On first contact with any customer, Otto states it’s an AI assistant. This is on by default and can’t be silently turned off — it aligns with CA SB 1001 and EU AI Act Article 50 norms.
It hands off instead of guessing
When Otto isn’t confident, it says so and routes to a human rather than fabricating an answer. “I don’t know” is an allowed — and expected — response.
You can pause or override anything
One switch pauses Otto entirely. Every draft is editable, every automation is toggleable, and you can take the wheel at any moment.
The honest limits
Every action Otto takes is on the record.
See the audit trail: a timestamped log of what Otto did, what it used, and who approved it.