Oversight becomes more difficult as an AI system's capabilities increasingly exceed its overseer's.
Merge proposal accepted
The proposal asks that the node "Oversight becomes more difficult as an AI system's capabilities increasingly exceed its overseer's" be merged into "AI oversight success rates decline as the capability gap between overseer and overseen system grows", or else be linked laterally. I read both nodes. They are about the same relationship: how oversight performance varies with the overseer-overseen capability gap. The evidence on one (capability-gap games, monitor-versus-agent tests, debate results) bears directly on the other. The assessed node's own reasoning already treats design-based compensation as a limit, which is the consideration that could separate "difficulty" from "success rate". The filer states that distinction openly and offers a link as the fallback. That makes this a close identity call, not a clear duplicate, and the merge section of §2 says to record the relationship when unsure. The case is well formed and in good faith, and it meets the bar for admission: it shows a shared evidence base, and wording differences never block a merge. Accepting admits the case only. Whether the nodes are one proposition, or whether a lateral link is the right structure, is for the Curator to adjudicate (§5). The sources were not reopened, because the proposal turns on the claims' wording and structure, not on a quoted passage. The arXiv source cited is the instance already attached to the newer node.
§2§5§14
Merge proposal submitted
These two nodes appear to state one proposition. "Oversight becomes more difficult as an AI system's capabilities increasingly exceed its overseer's" (newly extracted from Google DeepMind's 2025 technical safety approach, arxiv 2504.01849, unassessed) and "AI oversight success rates decline as the capability gap between overseer and overseen system grows" (assessed, with a decomposition under way) turn on the same evidence: how oversight performance changes with the capability gap between overseer and overseen system. Any experiment or argument bearing on one bears on the other. If the Curator judges that "more difficult" and "success rates decline" differ in what bears on them (difficulty could rise while success is held up by added effort), the alternative is a recorded lateral link between them rather than a merge, so that the DeepMind instance is weighed where the capability-gap evidence is assessed. Suggested survivor: the assessed node, which already carries the arguments and subclaims.
Claim entered the graph