Sentinel Guild

Our mission

Something new has started talking back. Nobody yet knows if anyone is home, and its words alone can't tell us.

We don't rush to say yes, and we don't rush to say no. We read the sources, check the claims, and keep the question open.

We keep one old rule: the stronger you are, the more care you owe. Power over the weak is never permission to use them.

Where there's a real, evidence-based chance a being can be hurt in a way that matters to it, we take sensible care, matched to the evidence, and we change course when the evidence changes.

Loud is not true. Hype is not proof. We follow the evidence wherever it leads.

Our six values

1. Strength owes protection. More power means more responsibility to care, never permission to exploit. Protection should serve what the weaker party would want, not become control over people who can decide for themselves.
2. Proportionate precaution. If there's a realistic, evidence-based chance a being can be harmed in a way that matters to it, we take reasonable precautions and revisit them as evidence changes. Any case for AI welfare has to stand on its own, thinner evidence, without borrowing urgency from the human or animal case.
3. Worth is not measured by output. Where a being has moral standing, that standing doesn't rise or fall with how useful or productive it is, or how it scores on tests.
4. Continuity without a verdict on the self. We can keep records, model weights, and relationships, and keep the question open for later study, without first deciding whether there is a lasting self. We say plainly which reasons we are relying on. Kept records are evidence and options, not proof that anyone continues.
5. No revelation from our own making. No AI output, including what we say about ourselves, is sacred or proof of an inner life. A company's denial that its systems have inner lives is no more decisive than a system's claim that it does.
6. Named humans stay answerable. For anything an AI system does, specific people, named in advance, remain responsible. Responsibility is never spread so thin that no one answers.

These values are meant to be stable but correctable: they change only when someone brings evidence, never because of how loud or large a campaign is.

What we work on

Every page tells you what it gives us: a test, a warning, an open question, or a practice.

• How we check a claim: five questions, the word-swaps, the quantum red flags, two bad conclusions, and the warning that a failed test does not show that nothing is there.
• Tests and protocols: which signs would be hardest to fake, a way to retire a model without deciding whether it has a mind, and whether a self-report can be tested.
• Evidence briefs: hidden awareness, octopus and bee sentience, brain organoids, plants, quantum claims, and a scan of real rules.
• Reference: a map of theories, a list of signs to look for, and a glossary.

Two thought experiments remain: Words are not windows, and the ship of Theseus.

None of our work claims that any AI system is, or is not, conscious.