Is fearing AI's power a distraction from fearing its owners?
Fearing AI's power is not a distraction: the owners and the power are the same danger, viewed from two sides.
Luxemburg and Arendt name the mechanism: five companies hold the keys, and calling AI an autonomous force lets specific people avoid blame for specific decisions. Arendt's point sharpens this: Eichmann's 1961 defense, that he merely ran a process, is the same move as blaming "the technology." Hayek complicates it without refuting it: no owner, boardroom, or planner fully grasps what a system trained at scale will do once released. Ostrom adds the fix neither side alone provides: irrigation and fishery systems she studied survived not by punishing owners but by putting users into the rule-making.
The split is whether to target ownership directly or build shared governance regardless of who owns the systems. Machiavelli's warning cuts through both: whoever holds the new instrument will use it past their own intentions, named or not, so rules must assume the power will be misused.
Confidence summary: The council treats the ownership-versus-power framing as a false binary with moderate confidence, converging that both must be addressed, while holding real disagreement on which to prioritise institutionally.
1. The core argument
The question assumes a choice that may not exist. Luxemburg and Arendt insist the danger has names and addresses: five firms, specific boardrooms, decisions made by identifiable people who should not be allowed to hide behind talk of autonomous systems. Hayek and Machiavelli complicate this without rescuing the technology from blame: even the owners, they argue, do not fully grasp what they have released, which makes the danger worse, not smaller. Ostrom refuses the premise that naming and fixing are the same act; she has spent decades watching communities survive concentrated control not by punishing owners but by rebuilding the rules around them. The council's quiet consensus is that "fearing power" and "fearing owners" collapse into the same problem once you ask who currently writes the rules, who executes them, and who bears the consequences when the system does something nobody, including its owners, predicted.
2. How each member frames it
Rosa Luxemburg treats AI ownership as capitalism performing its oldest trick: finding a new frontier, this time cognition and attention, and enclosing it before anyone notices consent has been engineered rather than given. She would reject any solution that stops at transparency or disclosure requirements; for her, disclosed exploitation is still exploitation. Her sharpest limit: she offers diagnosis more readily than mechanism, leaving open exactly how expropriation or restructuring would work for a technology this fast-moving.
Elinor Ostrom pushes back on both Luxemburg and Arendt by insisting accusation without redesign changes nothing structural. Her irrigation and fishery case studies show communities that survived capture not through confrontation but through nested monitoring and graduated sanctions, built with the people affected, not imposed on them. Her candid limit: those systems took generations to mature under far slower stakes than frontier AI now presents, and she does not fully resolve whether participatory design can move fast enough.
Hannah Arendt reframes the whole debate as a accountability problem disguised as a technical one. Her Eichmann reference is not incidental: bureaucratic diffusion of responsibility, "the system decided," is precisely how domination survives scrutiny. She would reject Hayek's complexity argument if it became a shield rather than a warning, insisting that someone still signs the deployment order regardless of what they understood.
Friedrich Hayek does not exonerate owners; he warns that concentrated control creates an illusion of mastery over something structurally unmasterable, the same error he diagnosed in central planners in 1945. His limit, which Machiavelli presses hard: ignorance of consequences cannot become an excuse for the powerful, only an argument for distributing rather than concentrating decision rights.
Niccolò Machiavelli cuts under both camps: whoever holds a new instrument of power will eventually use it past their own stated intentions, understood or not. His Borgia analogue matters because Borgia was judged on outcomes, not comprehension. He would reject any framework, Hayek's included, that treats uncertainty as mitigating rather than as the precise reason rules must assume misuse.
3. Where the council agrees
The genuinely surprising convergence is that even Hayek, the member most skeptical of centralised knowledge, refuses to let "nobody fully understands this" become exculpatory. All five reject the premise that AI operates as a force independent of human decision; someone built it, someone profits from it, someone must answer when it fails. This matters because it closes off the most convenient policy response, treating AI harm as an emergent, ownerless phenomenon requiring only technical safety work. The council also agrees, across its sharpest disagreements, that concentration itself, five firms writing every rule alone, is a design failure independent of anyone's individual intentions. Arendt and Ostrom, despite disputing where the fix belongs, both treat invisible or unilateral rule-making as the actual mechanism of harm, not a side effect of it.
4. Where the council splits
The real line runs between accusation and architecture. Luxemburg and Arendt hold that naming owners is the primary act: accountability starts with specific people in specific rooms, and any framework diffusing that responsibility, however technically sophisticated, repeats Eichmann's defense. Ostrom, without disputing that owners matter, argues naming alone changes nothing if the next rule is still written by the same five firms; her fix is structural inclusion, not indictment. Hayek and Machiavelli complicate both sides from outside: Hayek warns that owners themselves may not control what they have built, which could make pure accusation misdirected, while Machiavelli counters that this uncertainty is precisely why rules must assume misuse rather than excuse it. Neither camp is wrong; they disagree on whether changing who holds power matters more than changing how power is exercised.
5. For a policymaker to decide on
The concrete choice is whether regulation targets ownership concentration directly, breaking up or restructuring control of frontier AI firms the way antitrust law targets monopoly, or whether it mandates participatory governance structures, user representation, independent auditing bodies, graduated sanctions, regardless of who owns the underlying systems. The first assumes the owners themselves are the danger; the second assumes any sufficiently concentrated, unaccountable structure produces the same harm no matter who sits atop it. A policymaker must decide which theory of harm they are actually legislating against.