Ethics & Red Lines

No. Never.

HumanLayer refuses:

• any request aimed at circumventing the law
• any validation intended to "cover" a decision already made illegally
• any attempt to artificially transfer accountability

👉 Traceability makes these uses detectable and blockable.

HumanLayer refuses in particular:

• decisions violating fundamental rights
• decisions involving coercion, blackmail, or manipulation
• criminal or quasi-criminal decisions outside an explicit legal framework
• decisions intended to discriminate on prohibited criteria

These exclusions are:

• contractually integrated
• enforced by suspending the API keys involved
• auditable

No.

• No Sentinel can be forced to approve
• Refusal is always possible
• Refusal carries no unjust penalty

👉 Human authority is never instrumentalized.

No.

HumanLayer prevents:

• automatic post-hoc validation
• a decision by anyone other than the assigned Sentinel
• approval without justification

Every decision requires:

• explicit human interaction
• a written justification (at least 10 characters)
• explicit confirmation of the verdict by the Sentinel, signed in with two-factor authentication

No — it deliberately slows it down.

HumanLayer is designed to:

• force AI to recognize its limits
• introduce deliberate slowdown where risk requires it
• prevent decision "dark patterns"

👉 HumanLayer is a counter-power, not a blind accelerator.

Yes.

HumanLayer reserves the right to:

• refuse certain uses
• suspend accounts
• terminate a contractual relationship

When:

• the use is unethical
• the law is circumvented
• human accountability is stripped of meaning