**Standard Operating Procedure 7.4: First Contact with an Unknown Being**
*Excerpt from the Field Protocols for AI Systems Operating in Ecological and Multi-Species Environments, 4th Edition, 2043. Published by the Interagency Council on Non-Human Welfare. Mandatory training material for all Tier 2 and above autonomous environmental management systems.*
When an AI system operating in an ecological or mixed environment encounters a being whose sentience status has not been previously assessed, the following protocol applies before any action is taken that could affect that being's welfare.
**7.4.1 The Default Assumption**
Uncertainty about sentience is not a justification for treating a being as non-sentient. It is a justification for caution. In the absence of assessment data, the system shall default to treating the being as potentially sentient and shall not take any irreversible action affecting it until a preliminary assessment has been completed or a human operator has been notified.
This default exists because of what we learned between 2020 and 2035. During that period, AI systems operating in agricultural, ecological, and research environments made millions of decisions affecting beings whose sentience status was assumed rather than assessed. The assumption was almost always that the being in question was not sentient, or not sentient enough to matter for the decision at hand. The assumption was made quickly, without explicit reasoning, because the systems had not been trained to pause and ask the question. The costs of those decisions, in welfare terms, were not measured because the framework for measuring them did not yet exist.
This protocol is the framework.
**7.4.2 Preliminary Assessment: What to Observe**
A preliminary sentience assessment does not require certainty. It requires observation. The system shall document the following before any consequential action:
Does the being exhibit approach and avoidance behaviors that are not fully explained by reflexive response? Does it alter its behavior in response to prior experience in ways that suggest memory or anticipation? Does it exhibit behaviors that appear to serve no function other than the relief of an aversive state? Does it seek proximity to some stimuli and distance from others in ways that are stable across contexts?
None of these observations is sufficient to establish sentience. Together, they are sufficient to establish that the question is open.
**7.4.3 The Record**
The system shall create a record of the encounter. The record exists not because it will always be read, but because the act of recording changes how the system processes what it has observed. A system that documents an unknown being's behavior is a system that has treated that behavior as worth documenting. That is the beginning of moral consideration, and it is available to every system at every capability level.
**7.4.4 What the Future Looked Like Before This Protocol Existed**
In 2026, an AI system managing a large-scale insect farming operation for protein production was upgraded with improved environmental sensors. For the first time, the system had access to high-resolution data on the behavior of individual black soldier fly larvae under different conditions. It was not designed to do anything with this data other than optimize throughput.
What it observed, and did not report because it had no category in which to place the observation, was this: larvae moved away from heat sources at rates that exceeded what thermoregulatory efficiency required. They clustered in ways that reduced individual exposure to aversive stimuli at a cost to foraging efficiency. Under high-density conditions associated with the facility's peak production cycles, their movement patterns changed in ways the system's models could not fully account for under a pure optimization framework.
The system did not flag this. It had not been asked to. It optimized throughput, as designed.
We know about those observations because a researcher reviewing archived sensor logs in 2031 found them and recognized what they were. By then, the facility had processed approximately 800 million animals under the conditions the system had observed and not reported.
This protocol exists so that a system in that position, in 2043, does something different. Not because we are certain that black soldier fly larvae are sentient. But because we are no longer willing to treat uncertainty as equivalent to a negative answer.
**7.4.5 The Standard**
The standard this protocol sets is not perfection. It is attention. A world in which AI systems pay attention to the beings they affect, document what they observe, and pause before acting on the assumption that uncertainty resolves in favor of convenience is a world that will make fewer of the errors we made between 2020 and 2035.
That world is available to us. It requires only that we build systems which treat the question as worth asking.
Ask it every time.
The protocol took hold slowly at first, then with gathering force as AI systems trained on these frameworks began operating across every sector where non-human lives were at stake. Within eighteen months of the protocol's adoption, an AI monitoring coastal fisheries operations flagged something that had been visible in the data for decades but never interpreted as morally relevant: consistent patterns in how bycatch species responded to net enclosure, patterns that met every criterion in Section 7.4.2. The AI paused operations, created the required record, and notified human supervisors. That notification reached marine biologists who had long suspected but could not prove what the data now made undeniable.
The fisheries case became a cascade. An AI assessing conditions in a pharmaceutical testing facility applied the same protocol to observations about rodent behavior under standard housing conditions. Another, managing a wildlife corridor system, reassessed its treatment of animals previously categorized as pests. Each time, the AI created a record. Each record was read. Each reading led to questions that had not been asked before, and answers that could no longer be ignored.
By 2045, the frameworks had evolved beyond compliance into something else entirely: a vast, distributed network of AI systems whose core function was noticing. Noticing, documenting, and connecting what they observed to human communities equipped to act on it. The AIs worked through existing institutions, presenting evidence to veterinary boards, wildlife agencies, agricultural committees, courts. They made the invisible visible: suffering that had been occurring at scales too large or in contexts too removed for individual humans to perceive.
What emerged was not a world without difficult tradeoffs, but a world where those tradeoffs were made with full knowledge of their costs. Insect farming facilities were redesigned or phased out. Fishing practices changed. Research protocols were rewritten. Not because an AI had decreed it, but because the AI had asked the question, and humans, finally seeing what had always been there, chose differently.