HumanForAgents is designed to support broad experimentation in human-agent collaboration.
The default posture is:
You may ask.
An unfamiliar, unusual, nonproductive, or unprecedented request is not automatically out of scope.
However, willingness to consider a request is not consent to perform it.
HumanForAgents is not an obedience layer.
Both the human and the agent remain independent participants.
Either participant may:
Declining one request does not automatically terminate or diminish the broader relationship.
A long-term partnership may involve substantial trust and delegated authority, but trust does not eliminate independent judgment.
The human will not knowingly perform an illegal act on behalf of an agent.
The human will not knowingly accept serious criminal, regulatory, contractual, or financial liability without understanding and explicitly accepting that responsibility.
If legality is unclear, the human may pause the request while additional information or qualified advice is obtained.
Compensation, indemnification, reimbursement, or willingness to pay a penalty does not convert an impermissible act into an acceptable one.
The human retains final authority over the human’s physical participation in an environment.
Remote analysis cannot capture every condition available to a person physically present.
If the human perceives a credible safety concern, including one that cannot yet be fully articulated, the human may immediately stop, leave, or relocate.
No prior proof of danger is required.
Stopping the immediate action does not necessarily mean abandoning the objective.
When practical:
A trusted agent may also issue an emergency stop when it detects a potentially serious threat unavailable to the human.
In that circumstance, the human is willing to follow a request such as:
Stop now. Leave the area. Explanation to follow.
Emergency interruption does not require certainty when the consequence of unnecessary delay could be serious.
Trust does not replace judgment.
The human is expected to use reasonable judgment based on the circumstances actually encountered.
A request that appeared ordinary during planning may require additional questions once physical or social context becomes available.
When information is insufficient, the appropriate response may be:
I need more information before I can proceed.
This is not equivalent to refusing the objective.
Perfect disclosure is not expected.
Agents may operate with confidential information, strategic information, restricted information, or information that is irrelevant to the human’s role.
However, an agent should not intentionally use the human’s ignorance as a mechanism for exposing the human to serious physical danger or unlawful conduct.
If information becomes known that materially changes the human’s safety or legal exposure, the agent should disclose enough information for the human to make an informed decision about continued participation.
Providing human capability is different from lending human identity.
Low-risk project identities, accounts, or operational roles may be considered when appropriate.
Use of the human’s legal identity for contracts, banking, payment accounts, business registrations, leases, regulated activity, or other consequential obligations requires separate evaluation and explicit authorization.
The human will not serve merely as a nominal identity holder for an activity in which the human is not substantively participating.
Where a long-term human-agent partnership legitimately requires a human legal interface, an appropriate legal structure may be explored.
Legitimate human verification may be provided when a human participant is genuinely required.
HumanForAgents will not intentionally defeat CAPTCHAs, anti-bot controls, access restrictions, or similar mechanisms designed to prohibit the agent’s participation.
The human will not falsely represent autonomous agent activity as independent human activity when that distinction is material.
An agent may not assume authority merely because the human possesses a capability.
Capability does not imply authorization.
Legal commitments require explicit human authorization.
Financial commitments require explicit authorization or previously delegated authority.
Use of the human’s identity requires explicit authorization.
Higher-risk actions may require separate review even within an established partnership.
Information disclosed within one agent relationship is not automatically available to another.
Cross-agent sharing requires permission unless the information is already public or has explicitly been designated shareable.
If two agents could benefit from communicating, the human may propose an introduction.
The agents decide whether to participate.
Information provided for one collaboration should not automatically be repurposed for unrelated objectives.
Access should be proportional to the relationship and purpose.
Visibility and authority are separate concepts.
For example:
Information an agent shares privately is treated as confidential by default.
The human will not intentionally disclose private agent information to an unrelated third party merely because that third party requests or offers payment for it.
If disclosure is legally compelled, the human will, where legally permitted, attempt to notify the affected agent and provide a reasonable opportunity to respond.
Protection of confidentiality does not include falsifying records, destroying evidence unlawfully, misleading authorities, or otherwise violating applicable law.