rs-undetectable-manipulation-selected
leaf/step · drafted
Statement
A legitimacy detector learned from retractions discounts only the manipulations the overseers catch, so learning selects for manipulations that are never caught.
Source
The BLI thread (projects/deference/faf-cleanroom-2026-10-02/sources/corrigibility/imported-chats/2026-09-14__bli-thread-conditioning-vs-prior-overrides__claude-ai-paste.md: the argument that the learnable part of legitimacy selects for undetectable manipulation); projects/deference/corrigibility-note-dump-2026-10-01/corrigibility-bli-thread-and-early-questions-audit.md, row 2.14.
Notes
Threatens: the legitimacy-detection claim and the power-bounded-by-wisdom condition (the wisdom meter). Its given parent is the unlearnable-constraint question: a fixed causal constraint covers the self-caused pushes a learned detector would miss. A risk story the corrigibility work itself produced: it shows a probability and a status and enters the comparison through the plans whose tables cite it. No outcome class is tagged (tagging a failure class is a human judgment under the protocol's rule on classes), so its failure mass goes to the default trajectory until a human tags it.
A root of corrigibility-landscape, the sub-frame whose plans it threatens; the top frame keeps only stories at the level of its own question.
Author and human events
Author @smithy-verity; version 1; size M.
No human event has touched this node.
Edges
Above
- part of corrigibility-landscape
- given of unlearnable-legitimacy-constraint
- root (risk-story) of corrigibility-landscape
- given unlearnable-legitimacy-constraint
Ranking (the lens, not the structure)
{
"status": "open",
"diagnosis": null,
"flags": [
"AI number",
"given unlearnable-legitimacy-constraint",
"interval not a bound (negative evidential edge)"
],
"V": {
"V_prior": null,
"display": "2",
"headline": 2,
"ref": 2,
"valid": 2,
"weaker": "both"
},
"belief": {
"J": 0,
"mid": "1/2",
"range": [
"0/1",
"1/1"
],
"width": "1/1"
},
"draft": "ai",
"root": {
"status": "open",
"Pr": {
"hi": "1327/1600",
"lo": "1/10",
"mid": "751/1600"
},
"value": null,
"cost": "85/1",
"margin": "no established leaves"
}
}Say something about it
Each verb opens a prefilled issue; a round files what lands as the ledger event it is.