When AI Goes Too Far

The line is not where most founders think. It is not capability. It is consent.

What Looks Like Going Too Far

• Generating personalized content at scale.
• Using AI to draft customer emails.
• Replacing human roles with AI in the support flow.
• Letting AI write your blog.

None of these are the line.

What Actually Crosses the Line

• Misrepresenting AI as human. A chatbot that says "I am Sarah from support" with no disclosure.
• Training on customer data without consent. Whether contractually allowed or not, the trust cost is real.
• Generating content that looks like a customer wrote it for marketing purposes.
• Outputs that confidently misinform in stakes-bearing contexts (medical, legal, financial) without guardrails.
• Removing human escalation paths for users who need them.

Why Consent Is the Right Frame

Capability arguments age badly. "AI shouldn't generate email" sounds quaint already. Consent arguments don't age. The customer agreeing to AI in their workflow is the durable line.

How to Stay on the Right Side

• Disclose AI in customer-facing flows.
• Make the human escalation one click.
• Get consent before training on customer data.
• Add guardrails to high-stakes outputs.
• Audit your AI use semi-annually for drift.

The Honest Take

You can ship aggressive AI features and stay on the right side of this line. The founders who don't are usually the ones who skip the disclosure, the escalation, and the consent. None of those are hard. They are just easy to skip.