M3-A1
Local governance remains necessary → but governance within one domain does not ensure adequate coverage of the whole
Governance specific to military affairs, cyberspace, biology, economics, law, platforms, and other fields remains indispensable. The problem is that when risks couple across domains, optimizing one domain may shift costs to another or miss feedback across systems.
Governance-level assessment
Boundary: M3 cannot reject professional governance in the name of overall risk. If cross-domain coupling is weak, interfaces are clear, and feedback is adequate, domain-specific governance may already suffice.
M3-A2
Multiple risk domains change together → cross-domain coupling and feedback loops intensify
As a general capability, AI enters interstate competition, critical infrastructure, cyberspace, biology, cognition, and organizational decisions simultaneously. These cease to be merely parallel risks: an action in one field may alter threat perceptions, resource allocation, and action thresholds in another.
Structural assessment
Boundary: The extent of coupling must be demonstrated through specific interfaces and outcomes. Shared use of AI alone does not establish a systemic feedback loop.
M3-A3
Cross-domain coupling intensifies → governance feedback cycles may lag behind risk iteration cycles
When model capabilities, deployment patterns, attack paths, autonomy, and organizational adoption change faster than regulation, interagency coordination, incident reviews, and international negotiations, systems may enter their next state while governance is still understanding the previous problem.
Assessment of temporal structure
Boundary: Do not apply a universal monthly or quarterly threshold. The actual criterion is whether changes in key risks consistently outpace the identification–authorization–intervention–review loop.
M3-A4
Delayed feedback + mutual suspicion → local rationality may jointly generate greater insecurity
A state, institution, or platform may accelerate, add redundancy, pre-authorize action, close supply chains, or expand autonomy to reduce its own vulnerability. Under mutual suspicion, others may interpret these moves as threats and respond symmetrically or asymmetrically.
Strategic-interaction mechanism assessment
Boundary: Local rationality does not necessarily worsen overall outcomes. Transparency, verification, hotlines, shared guardrails, reversible authorization, and stable interdependence may produce the opposite result.
M3-A5
Local rationalities amplify one another → the probability of irreversible or hard-to-reverse consequences rises
Where actions involve autonomous weapons, critical infrastructure, biological risks, nuclear deterrence, social trust, or large-scale cognitive manipulation, a single error may cross the available window for remedy. The loss is then no longer merely a recoverable local incident.
Assessment of risks with major consequences
Boundary: Distinguish genuinely irreversible outcomes, outcomes difficult to reverse over long periods, and outcomes reversible at high cost; do not call every severe consequence irreversible.
M3-A6
Greater irreversibility → the old civilization's safety margin for recovering from mistakes shrinks
Many past wars, institutional failures, and technological accidents imposed enormous costs, yet humanity could recover through time, rebuilding, and institutional correction. If actions with major consequences become faster, broader, harder to attribute, and harder to recall, that margin for recovery may decline substantially.
Assessment of historical conditions
Boundary: This does not absolutize history as a past that always allowed another chance and a future that never will. Judge specific risks by recoverability, alternative paths, and reconstruction costs.
M3-A7
A shrinking safety margin → the sufficiency of gradual improvement alone comes into question
If old governance relies mainly on discovering problems afterward, handling them within separate domains, and repairing them gradually, while risk generation, cross-domain feedback, and irreversibility continually exceed its capacity to close the loop, continued gradualism may remain valuable without sufficing to guarantee overall safety.
Necessity test
Boundary: M3 questions only whether gradualism is sufficient, not gradual reform itself. If old governance can restore safety margins through faster feedback, cross-domain coordination, and reversible design, the necessity of a civilizational leap is weakened.
M3-A8
The sufficiency of gradualism comes into question → we must examine how risks are generated, not only their consequences
When risks arise repeatedly from ways of pursuing interests, competitive structures, power relationships, and insecurity, rather than from one technical defect, the next step must move from repairing tools to changing mechanisms of risk generation. M4 therefore turns to upgrading how self-interest is realized.
Bridge between nodes M3 → M4
Boundary: Proceed to M4 only when continuing regeneration of risk through structural motivations is supported. If the problem is mainly a single technical defect or execution error, prioritize technical or institutional repair without necessarily elevating it to civilization's underlying logic.