Law · Industry Pulse Editorial Team · · 5 min read
The Disposable Subsidiary: How Big Tech Is Bankrupting Its Way Out of AI Liability
Faced with existential copyright judgments, generative AI companies are reviving a controversial legal maneuver to trap their toxic liabilities in shell companies designed to die.

The corporate filing in Wilmington, Delaware, was remarkably brief. A prominent artificial intelligence developer, facing a mounting cascade of copyright infringement and algorithmic defamation lawsuits, had quietly executed a corporate restructuring. The parent company retained the user interface, the enterprise contracts, and the cash flow. A newly formed subsidiary was handed the foundational model’s training data, the underlying weights, and the entirety of the legal liability.
It was a classic "divisive merger," a maneuver historically favored by industrial conglomerates choking on asbestos claims. Now, in late 2026, it is becoming the primary legal architecture of the generative AI boom.
For the past three years, the tech industry has treated the legal risks of artificial intelligence as a lobbying problem. Executives testified before Congress, begging for safe harbors and regulatory moats. They argued that holding them liable for the ingestion of copyrighted material would stifle American innovation. But while the lobbyists stalled, corporate lawyers were building a firewall. They realized that you do not need to win the copyright debate if you can structurally guarantee that the entity losing the lawsuit cannot pay the judgment.
Welcome to the Algorithmic Two-Step.
The mechanism borrows heavily from the "Texas Two-Step," a controversial bankruptcy maneuver previously utilized by companies facing mass torts, such as Johnson & Johnson with its talcum powder litigation. The premise is brutal but legally elegant: a solvent company splits itself into two. One entity gets the valuable assets; the other gets the toxic liabilities and just enough funding to theoretically settle claims. The burdened entity immediately files for Chapter 11 bankruptcy. The solvent entity then asks the bankruptcy court to halt all litigation against the entire corporate family.
Applying this to artificial intelligence requires a novel interpretation of intellectual property and corporate structure. The core problem with an AI model is that its liability is baked into its architecture. If a model was trained on billions of unlicensed images or texts, the infringement is not a single event; it is the very foundation of the product. You cannot simply delete the offending data without destroying the model through catastrophic unlearning.
Therefore, tech companies are separating the ownership of the model from the application of the model.
Under these new structures, the "Bad Co" subsidiary owns the neural network and the training apparatus. It is the entity that scraped the internet, ingested the copyrighted books, and processed the private data. It absorbs all the legal risk. The "Good Co" parent company operates as a mere customer. It licenses the API from its own subsidiary. It builds the consumer-facing chatbot, secures the lucrative enterprise subscriptions, and collects the revenue.
When the inevitable multi-billion-dollar judgment arrives—say, a class-action victory by publishers or a massive defamation award—the plaintiffs will find their victory hollow. Their judgment is against the Bad Co. The Bad Co has no enterprise revenue, no cash reserves, and no physical assets beyond a cluster of depreciating servers. It will file for bankruptcy, trapping the plaintiffs in years of restructuring negotiations where they will fight over pennies on the dollar. Meanwhile, the Good Co continues to operate, its revenue untouched, claiming it merely licensed a tool from a vendor that unfortunately went under.
This is the financialization of legal liability. It treats copyright infringement not as a moral failing or a legal prohibition, but as an unquantifiable debt that can be securitized, isolated, and discharged.
Plaintiffs’ attorneys are rapidly adjusting their tactics, attempting to pierce the corporate veil. They argue that these subsidiaries are undercapitalized shams, completely controlled by their parent companies. They point out that the engineers writing the code often share the same cafeteria, the same stock options, and the same executive leadership.
Courts are currently struggling with the concept of "algorithmic piercing." Corporate separateness usually depends on financial independence—separate bank accounts, distinct boards of directors. But how do you evaluate separateness when the asset in question is a mathematical representation of human knowledge? If the parent company provides the prompts and the subsidiary provides the outputs, who is actually committing the infringement?
Some federal judges in the Third Circuit have signaled a willingness to look past the corporate shell, noting that the Good Co's entire business model relies on the Bad Co's alleged theft. Yet, bankruptcy courts remain notoriously friendly to debtors. The fundamental purpose of Chapter 11 is to preserve the going concern of a business, resolving claims collectively rather than letting a race to the courthouse dismember the company. When an AI subsidiary files for bankruptcy, the court's priority is organizing the chaos, a process that naturally shields the parent company from immediate discovery and trial.
The insurance industry is quietly accelerating this trend. Two years ago, securing a comprehensive liability policy for a frontier AI model was nearly impossible. The underwriting risk was too opaque. Today, insurers are offering specialized policies specifically designed to fund the bankruptcy trusts of these sacrificial subsidiaries. By capping their exposure to the exact amount the subsidiary is capitalized for, insurers can finally price the risk. They are not insuring against copyright infringement; they are insuring the bankruptcy process itself.
There is a profound economic consequence to this legal engineering. By capping downside risk through disposable subsidiaries, the law is actively incentivizing reckless data practices. If a company knows that the worst-case scenario is the bankruptcy of an isolated shell company, the rational move is to scrape everything, train on everything, and ignore the intellectual property rights of creators entirely. The market rewards the most capable models, and capability requires data. The penalty for stealing that data is simply the cost of organizing a Chapter 11 filing.
We are witnessing the creation of a shadow legal system for the digital economy. The traditional mechanisms of deterrence—injunctions, statutory damages, punitive awards—are failing because they were designed for unitary corporations that had something to lose.
The ultimate takeaway for corporate America is that legal exposure is no longer a barrier to technological deployment. It is merely an accounting variable. The companies that will dominate the next decade of artificial intelligence are not necessarily those with the most brilliant researchers or the most efficient compute clusters. They are the ones with the most ruthless corporate structuring, capable of generating billions in value while systematically bankrupting their mistakes.
Staff analysis by the Industry Pulse newsroom, edited by Stepan Kroka, under our editorial policy.
More Analysis
Law · Industry Pulse Editorial Team
The Neural Subpoena: Why Corporate Litigation Is Targeting Employee Brainwaves
Hospitality · Industry Pulse Editorial Team
The Acuity Premium: Why Luxury Hospitality Is Evolving Into Outpatient Infrastructure
Sports Industry · Industry Pulse Editorial Team
The Varsity Arbitrage: Why Wall Street is Buying Up the Youth Sports Economy
Healthcare · Industry Pulse Editorial Team
The Micro-Foundry Premium: Why America’s Hospitals Are Printing Their Own Pharmaceuticals
Finance · Industry Pulse Editorial Team
The Obsolescence Swap: Why Wall Street is Trading the Half-Life of Silicon
Real Estate · Industry Pulse Editorial Team
The Hydrological Premium: Why Wall Street Is Pricing Land by the Gallon