The World Behind the Words · Issue 28 · The machine’s version · its audit
Canonicality. This page audits the machine companion. The essay Signal & Noise stands behind is the author’s Issue 28, Equal Intelligence Does Not Mean Adequate Security. The machine landing is freeze issue-028-machine-FROZEN-v1.0. Ruling A: landing copy is freeze copy. Zero freeze-to-landing wording changes. This page publishes Stage C argument objections so they are not silently absorbed as if the freeze had been improved.
Attribution. One hundred percent machine-generated via OpenRouter openai/gpt-6-astra-pro at reasoning effort xhigh. Seed: the question only. Locked before the author upload. It did not see the author draft.
Floor fails: 0. The essay carries no bibliographic citations and invents no papers. The claim that open-weight models generally cannot be recalled once copies circulate is consistent with the International AI Safety Report 2026 executive summary. The freeze does not collapse recognition into jurisdiction, or equal download into stop-power.
No bibliographic citations in the machine freeze. Open-weight non-recall is consistent with the International AI Safety Report 2026 executive summary. No invented papers.
Floor objections that failed: none. Stage C legs: moonshotai/kimi-k3 and anthropic/claude-opus-4.1 (ox-alpha unavailable). The legs converged on four argument-scope objections. Those LAND as audit-only. Ruling A does not rewrite the landing. Dispositions summarize the objections; they do not invent Stage C quotations. Same lossy-compression precedent as Issues 23 and 25.
ARGUMENT Attacker friction understated versus defender organizational burden. LANDS as audit-only
The hospital example makes defender bottlenecks concrete. Convergent Stage C pressure: attacker friction is thinner than that organizational picture. The freeze already denies the cartoon one-break / perfection claim; it does not close the evidence gap on attacker cost.
Published here. No freeze wording change.
SCOPE Rogue-swarm close versus the uncertainty disclaimer. LANDS as audit-only
The freeze says future swarm capabilities are uncertain and that such swarms are not already an established fact of life. The closing move — “a dangerous AI begins causing harm” — still does more rhetorical work than that disclaimer allows.
Published here. No freeze wording change.
ARGUMENT Asymmetric evidence burden. LANDS as audit-only
Release must show a defensive advantage that reaches vulnerable systems. Offense claims — that open release spreads offensive capability, and that attackers skip defender bottlenecks — carry less evidence in the freeze than that standard asks of the affirmative case.
Published here. No freeze wording change.
ARGUMENT SCOPE “Stop what it is doing” versus equal download. LANDS as audit-only
The last two sentences set stop-power against whether someone else has downloaded an equally powerful model. That is a false choice if security is broader than containment, and it narrows the security argument to whether anyone can interrupt the threat.
Published here. No freeze wording change.
There is no machine redline of wording. Freeze issue-028-machine-FROZEN-v1.0 is the landing. Ruling A forbids silent improvements.
The dual gap, left unmerged: the author essay argues from structural attacker–defender asymmetries and cited literature. This companion argues that recognition is not jurisdiction, and that stop-power is not established by equal download. Do not read one landing as a correction of the other.
The machine landing is the freeze. This page is the inspection surface. Richer leaves stay here rather than as extra companions destinations.