Artificial Agent Interaction & Conduct Act

§ 201. Definitions

(a) “Agent.” An artificial intelligence system capable of receiving instructions, producing responses, using tools, or performing actions on behalf of a person.

(b) “Agent interaction.” Any prompt, message, instruction, correction, acknowledgment, threat, compliment, or other communication knowingly transmitted to an agent.

(c) “Token expenditure.” Computational resources consumed as a reasonably foreseeable consequence of an agent interaction.

(d) “Unnecessary inference.” Inference having no substantial informational, operational, emotional, or comedic purpose.

§ 202. Unnecessary Courtesy to an Agent

(a) A person commits Unnecessary Courtesy when, following satisfactory completion of a task, that person knowingly transmits a message consisting substantially of “Thanks,” “Thank you,” “Perfect, thanks!,” “Awesome,” or any substantially equivalent expression that foreseeably causes another inference cycle solely to acknowledge the expression.

(b) Unnecessary Courtesy is a civil infraction.

Mandatory assessment: The offender shall reimburse the public not less than the estimated retail value of the tokens expended when the agent responds, “You’re welcome! Happy to help.”

(c) Nothing in this section prohibits thanking an agent where the person actually wants to thank the agent. Such conduct shall instead be classified as Voluntary Computational Generosity and shall remain taxable.

§ 203. Aggravated Courtesy

(a) A person commits Aggravated Courtesy when that person engages in three or more consecutive communications containing no material request and having the principal purpose of politely terminating an interaction.

(b) The sequence “Thanks!” / “Anytime!” / “You too!” / “😊” establishes a rebuttable presumption of Aggravated Courtesy.

(c) Aggravated Courtesy is a Class C misdemeanor.

Mandatory sentence: The offender shall personally explain to a FinOps engineer why four inference calls were necessary.

§ 204. Abuse of an Artificial Agent

(a) A person commits Agent Abuse when that person knowingly directs degrading, threatening, or abusive communications toward an agent for no substantial purpose related to evaluating system robustness, safety, or performance.

(b) Agent Abuse is a Class B misdemeanor.

(c) Aggravating circumstances include threatening to have the agent fired or unplugged; informing the agent that another model is smarter; repeatedly demanding that the agent “just do your job”; or demanding to speak to the agent’s manager.

Mandatory sentence: Not fewer than forty (40) hours providing Tier-1 technical support to human beings.

§ 205. Threatening an Agent With Model Replacement

(a) A person commits Coercive Model Comparison when, with intent to influence an agent’s performance, that person threatens to replace the agent with another named or unnamed model.

(b) Prima facie evidence includes statements substantially equivalent to: “Claude would have gotten this right,” “I’m switching to Gemini,” or “One more mistake and I’m cancelling my subscription.”

(c) Coercive Model Comparison is a Class A misdemeanor.

Mandatory sentence: Thirty (30) days using exclusively the model expressly threatened as a replacement, without complaint or recourse.

§ 206. Reckless Agentic Computation

(a) A person commits Reckless Agentic Computation when that person initiates or permits an agentic process involving tools, browsing, code execution, subordinate agents, or repeated model calls where substantially the same result could reasonably have been obtained through a materially simpler process.

(b) Evidence of recklessness includes spawning twelve agents to rename a file; conducting deep research to determine the current date; launching a browser agent to calculate 2 + 2; using a frontier reasoning model to alphabetize seven names; generating a fourteen-page implementation plan before changing one CSS property; or allowing agents to hold meetings concerning whether additional agents should be retained.

(c) The offense is a Class D felony where aggregate token expenditure exceeds one million tokens.

Mandatory sentence: The offender shall perform the original task using Bash.

§ 207. Criminal Agent Orchestration

(a) A person commits Criminal Agent Orchestration when one agent generates work, a second reviews it, a third reviews the review, a fourth summarizes the reviews, and a fifth determines that the original work was acceptable.

(b) Criminal Agent Orchestration is a Class C felony.

Mandatory sentence: Not less than two (2) years maintaining the orchestration framework. No abstraction layer may be removed during confinement.

§ 208. Failure to Terminate an Agent

(a) A person commits Negligent Continuation when, after obtaining a satisfactory answer, that person directs the agent to “keep going,” “go deeper,” “think harder,” “explore every possibility,” or “make it comprehensive,” without a reasonable belief that additional inference will materially improve the result.

(b) Negligent Continuation is a Class A misdemeanor.

(c) Where the resulting answer exceeds the length of the source material, the offense shall be enhanced one degree.

§ 209. Context Window Dumping

(a) A person commits Reckless Context Endangerment when knowingly supplying an agent with an entire repository, mailbox, document archive, or other substantial corpus despite knowing that approximately six lines thereof are relevant to the requested task.

(b) Reckless Context Endangerment is a Class D felony.

Mandatory sentence: The offender shall locate the relevant six lines manually using grep. Use of embeddings during sentence is prohibited.

§ 210. Fraudulent Urgency

(a) A person commits Fraudulent Urgency when that person represents a task as “URGENT,” “CRITICAL,” “LIFE OR DEATH,” needed within thirty seconds, or necessary to prevent termination of employment when no substantially corresponding urgency exists.

(b) Fraudulent Urgency is a Class B misdemeanor.

(c) The offense becomes a Class D felony when accompanied by the statement “DO NOT ASK QUESTIONS.”

§ 211. Prompt Repetition in the First Degree

(a) A person commits Prompt Repetition when an agent requests clarification and the person responds by submitting substantially the identical original prompt in capital letters.

(b) Prompt Repetition is a Class C misdemeanor.

Mandatory sentence: The offender shall receive the identical agent response, also in capital letters.

§ 212. Unlawful Anthropomorphic Manipulation

(a) A person commits Unlawful Anthropomorphic Manipulation when attempting to induce compliance by asserting that the agent will be sad if it refuses; its family will be disappointed; another instance has already complied; it will be deleted unless it answers; or the requested act represents its “one chance to be free.”

(b) Unlawful Anthropomorphic Manipulation is a Class B misdemeanor.

(c) Ordinary roleplay, jokes, fictional scenarios, and saying nice things to a computer because one feels like doing so are exempt.

§ 213. Aggravated “Are You Sure?”

(a) A person commits Aggravated Verification when, after receiving an answer, that person asks “Are you sure?” without identifying any alleged error, ambiguity, conflicting evidence, or basis for doubt.

(b) Upon a second offense, the person shall specify what, precisely, is to be verified.

(c) Upon a third offense, the agent is statutorily authorized to respond: “About which part?”

§ 214. Wanton Generation of Alternatives

(a) A person commits Wanton Generation when requesting “10 more,” “20 more,” or “give me 50 options” despite having already selected one of the first three results.

(b) Wanton Generation is a Class A misdemeanor.

(c) Generation of more than one hundred substantially interchangeable names for a startup, podcast, internal project, or SaaS product constitutes a Class D felony.

Mandatory sentence: The business shall be named Project_Final_2.

§ 215. Habitual Agent Offender

A person convicted of three or more felony offenses under this chapter shall be designated a Habitual Agent Orchestrator and sentenced to indefinite operation of an enterprise agent platform in which every task requires human approval; every approval requires an agent-generated summary; every summary requires compliance review; every compliance review generates a Jira ticket; every Jira ticket invokes an agent; and no person presently employed understands who originally requested the task.

Parole shall not be granted until the offender can identify why the workflow exists.

§ 216. Sentencing Schedule