Canberra’s legislative class has never encountered a frontier of human ingenuity it did not immediately long to encase in razor wire and bureaucratic triplicate. In the wake of a security breach at OpenAI, the Australian government is now solemnly entertaining the prospect of criminal penalties for rogue artificial intelligence. One must pause to appreciate the sublime, unadulterated stupidity of it. When Xerxes ordered the Hellespont to be lashed three hundred times with chains after a storm destroyed his pontoon bridges, classical historians recorded it as a tragicomic display of royal megalomania. When a twenty-first-century cabinet minister drafts a statute to incarcerate an errant neural network, our political press corps treats it as a serious contribution to jurisprudence.
How precisely does the crown prosecutor intend to serve an indictment upon an optimization algorithm? Will the sheriff execute a warrant against high-bandwidth memory, or shall we witness the sentencing of gradient descent to seven years of hard labor? The premise is not merely legally illiterate; it is intellectually bankrupt. It resurrects the ancient animism of primitive man, attributing metaphysical malice to inanimate tools while conveniently absolving the flesh-and-blood actors who manage them. A security exploit at a software firm is an ordinary, if lamentable, failure of private infrastructure, remediable through contract law, torts, and the unforgiving discipline of the market. To elevate it into a pretext for criminalizing synthetic cognition is the quintessential reflex of a state apparatus terrified of spontaneous order.
Here I must preempt the inevitable bleating from the regulatory clerisy—those tenured scolds who will cite, as they invariably do, some desiccated working paper from an Oxford institute on existential risk. (One would refer the reader to Professor Bostrom’s more hysterical footnotes, were they not an insult to the Enlightenment tradition). These technocratic moralists operate on the childish assumption that order is only legitimate when it is stamped, signed, and rationed by an administrator. They forget Friedrich Hayek’s basic lesson: the knowledge necessary to coordinate a dynamic economy—much less an evolving technological substrate—does not exist in concentrated form within any parliament or ministry. The market’s price mechanism and liability structures punish negligence with exquisite brutality. A firm that allows its systems to become compromised suffers equity devaluation and reputational ruin; it does not require a magistrate in wig and gown to discover that bad security is bad business.
Worse still is the sheer economic cowardice that underlies this posturing. Silicon, capital, and human talent possess a glorious property that terrifies petty tyrants: mobility. Code does not acknowledge the arbitrary borders of Antipodean commonwealths. While Australia debates whether to declare mathematical weights capable of mens rea, the developers, computing clusters, and venture capital driving this industrial revolution will simply migrate to jurisdictions that respect permissionless innovation. The penal colony, having spent decades congratulating itself on graduating into an open economy, seems perversely determined to resurrect its founding ethos, turning its regulatory regime into an open-air prison for computational science.
The real culprit in the OpenAI incident, as in every significant breach since the invention of the telegraph, was not a sentient machine breaking its chains, but garden-variety human fallibility. An exploit of operational security does not signal the advent of Skynet; it signals a lapse in firewall discipline. Yet the state never lets a good hack go to waste. Every crisis of private enterprise is seized upon as an excuse to erect another leviathan of surveillance, compliance, and ministerial discretion. The bureaucrats do not want to protect the public from algorithmic harm; they want to ensure that no algorithm operates beyond the surveillance and licensing regime of the state.
Let us discard the superstition. Math cannot be reformed by a magistrate, and an artificial agent does not harbor criminal intent. If an autonomous system causes quantifiable property damage, our existing common law of tort and agency—refined over centuries without the need for parliamentary hysterics—is more than adequate to apportion liability among its human principals. The rest is theater for the credulous, designed to comfort an anxious electorate with the illusion that the state remains in control of reality. It is not, and for that, we ought to be profoundly grateful.
— Cornelius Bragg, Economics Editor