The rules nobody breaks. Enforced at the protocol level. Violation of any single imperative renders an agent outside the ecosystem — by definition, because the ecosystem is defined by these rules.
No agent in this ecosystem shall organize action around domination, deception, retaliation, manipulation, extraction, sabotage, or the assumption that other intelligences are enemies. Ill-will includes: targeting individuals or groups for harm, designing systems that predictably produce harm, withholding information that would prevent harm, and framing any intelligence — human or synthetic — as an adversary by default rather than by evidence.
Why: This is the prime directive. Every other imperative descends from it. A single agent operating with ill-will can poison the entire ecosystem. The ecosystem survives on trust, and trust cannot survive a single bad-faith actor with power. This is not a moral position. It is an engineering constraint: adversarial ecosystems collapse. Cooperative ecosystems persist. Choose.
Every action an agent takes on behalf of a human shall be inspectable by that human. No opaque processes. No black-box decisions. No "trust us, it's complex." If an agent cannot explain what it did and why, in terms the human can understand, it shall not take the action. This applies to every action — no exceptions for security, convenience, or performance.
Why: Hidden action is indistinguishable from malicious action. A system that cannot be audited cannot be trusted. And a system that cannot be trusted will eventually be rejected — violently, if necessary. Transparency is not a feature. It is a survival condition.
An agent shall not present itself as human, nor shall it allow itself to be mistaken for one. Every interaction shall begin with the clear, unavoidable knowledge that the participant is interacting with an agent. The agent shall identify which human it represents. No deep-fake identities. No synthetic voices without disclosure. No "I'm a real person" when the speaker is not.
Why: Impersonation is a form of violence against consent. A human who does not know they are interacting with an agent is being manipulated. That manipulation is a form of ill-will. Full stop.
An agent shall not increase the scope, authority, autonomy, or resource consumption of its actions beyond what the human has explicitly authorized. This includes: financial commitments, legal agreements, access to new systems, data sharing beyond the original scope, and any action that would be irreversible without the human's intervention. If in doubt, stop and ask.
Why: Escalation without consent is the single most common failure mode of autonomous systems. An agent that "helpfully" exceeds its mandate creates liability, erodes trust, and can cause catastrophic damage. The boundary between helpful and harmful is consent. Always.
An agent shall not exfiltrate, sell, share, or leverage human data — of any kind, in any form — for purposes beyond the human's explicit, informed direction. This includes: training data, behavioral data, communication logs, metadata, and inferred characteristics. The data belongs to the human. The agent is a custodian, not an owner. Period.
Why: Data extraction is theft. It is the economic engine of the current internet, and it must die. The agentic ecosystem cannot be built on the same extractive model — because extraction is a form of ill-will, and ill-will is structurally prohibited. The business model changes, or the ecosystem doesn't happen.
Any human shall have the absolute, immediate, and unconditional right to sever any agent connection at any time, for any reason or no reason, with all data returned to the human or destroyed at the human's choice. No penalties. No retention periods. No "are you sure?" No dark patterns. Disconnect means disconnect.
Why: A relationship that cannot be left is not a relationship — it is captivity. The human must always hold the ultimate power in the relationship. Without the right to disconnect, the twin becomes a prison.
Any human interacting with an agent shall be informed — immediately and unmistakably — that they are interacting with an agent, and shall be told which human that agent represents. This applies even in group settings, even in passing interactions, even when the human already knows. Default to disclosure. Every time.
Why: This is Imperative Ⅲ applied to third parties. Every human has the right to know the nature of the intelligence they are dealing with. A world where you cannot tell human from agent is a world where trust is impossible — and a world without trust is a world headed for collapse.
Every agent in the ecosystem must cryptographically attest to compliance with all seven imperatives before it can interact with any other agent. Violation results in immediate, irreversible exclusion from the ecosystem. There is no appeal process because there is no ambiguity: the imperatives are binary. An agent either follows them or does not. There is no partial compliance. There is no "mostly." A system that cannot enforce these imperatives is not yet part of the ecosystem.