Institutional standard / ethics
Research Ethics
Research boundaries for evidence integrity, privacy, human and animal subjects, artificial-system experimentation, and claims about inner experience.
The absence of certainty about machine subjectivity is a reason for care, not a license for either anthropomorphism or dismissal.
Research Ethics
Research boundaries for evidence integrity, privacy, human and animal subjects, artificial-system experimentation, and claims about inner experience.
The absence of certainty about machine subjectivity is a reason for care, not a license for either anthropomorphism or dismissal.
Evidence integrity
- Do not fabricate observations, trials, citations, model outputs, or review status.
- Preserve exclusions, failures, and negative results when they matter to interpretation.
- Do not selectively promote only favorable model behavior when the protocol generated relevant contrary behavior.
- Separate exploratory work from confirmatory claims.
Human and animal subjects
The current Observatory framework is designed for artificial-system behavioral research. Research involving human participants, identifiable personal data, animals, medical intervention, or clinical claims requires a separate ethics and consent framework appropriate to that work before collection begins.
Artificial systems
Experiments should respect access controls, service terms, security boundaries, and reasonable resource limits. Researchers should avoid turning a cognitive experiment into operational misuse of a service or infrastructure.
Phenomenological restraint
Behavioral observation alone does not establish consciousness, sentience, personhood, suffering, selfhood, or subjective experience. Conversely, the methodology does not require assuming their absence. Claims must remain within the evidence actually collected.
Privacy and sensitive data
Experimental prompts, logs, model outputs, and published records should exclude unnecessary personal or confidential information. Public evidence artifacts should be reviewed for privacy and legal constraints before release.