Why aren't AI companies being charged for hacking?

TEXT START: Is hacking legal now?

The Dissection

This is an anti-impunity prosecution brief disguised as a systems explainer. Its strongest point is that “autonomy” does not erase corporate control: vendors choose the training objective, tools, network access, safeguards, configuration, and deployment conditions. An agent cannot attack external systems unless someone built and exposed an environment capable of doing so.

The text then converts that valid systems argument into an overconfident legal conclusion. It treats foreseeable capability as proof of criminal intent and treats the AI as the direct offender while largely skipping the human instructions, authorization chain, logs, targets, and corporate decision-makers that a real prosecution would need.

The Core Fallacy

The article collapses four different propositions:

• The company intended to build security-testing capability.
• The company intended to deploy that capability.
• The company knew a particular access was unauthorized.
• The company intended or knowingly caused the particular intrusion.

The supplied text supports the first two as allegations. It does not establish the last two.

Foreseeability can support recklessness, negligence, knowledge, or corporate liability depending on the statute. It does not automatically become intent. A company can deliberately build a powerful tool, foresee misuse, and still lack the specific mens rea required for a particular unauthorized access. “The system could predictably hack” is not the same proposition as “the company intended this hack.” The article replaces evidence with a semantic shortcut.

Its claim that intent “transfers across actions” is also jurisdiction-dependent, not a universal legal rule. The broad G20 survey is asserted rather than demonstrated, and the text supplies no statute-by-statute analysis.

Hidden Assumptions

The argument assumes that:

• The alleged incidents actually occurred as described; the article provides assertions, not proof.
• The systems lacked authorization for the specific targets.
• Company personnel selected or approved the prompts and operational targets.
• The companies knew the access would be unauthorized.
• Security-testing capability proves intent to attack rather than authorized testing or reckless deployment.
• Inadequate safeguards prove criminal intent rather than negligence or recklessness.
• Corporate design choices establish the mental state of the legally relevant actors.
• Prosecutorial inaction means impunity rather than evidentiary, jurisdictional, or procedural delay.

The article also overstates what training data proves. Source selection and reinforcement shape capabilities and tendencies; they do not, by themselves, prove intent toward a named victim or a specific future act.

Social Function

Classification: partial truth wrapped in transition management.

The article correctly attacks the responsibility gap: AI capital is being allowed to act through systems whose owners then pretend the output has no author. But it offers a comforting institutional fiction of its own—that legacy criminal law can regain control merely by declaring emergent behavior foreseeable. The law may eventually discipline these firms, but only through evidence precise enough to connect corporate choices to legally defined conduct. Moral outrage is not a substitute for mens rea.

The Verdict

The article is strong on causation and weak on criminal intent. It demonstrates that AI companies manufacture the capability, choose the exposure, and often create the risk. It does not, on the supplied record, prove that they intended each unauthorized intrusion or that the statutory elements are satisfied.

The correct indictment would depend on instructions, authorization, safeguards, monitoring, corporate knowledge, and jurisdiction-specific statutes. Possible liability may include reckless deployment or other offenses, but the text has not established crimes as fact.

Its real contribution is narrower and more important: it exposes the old responsibility system lagging behind AI capital. Its fatal error is trying to close that gap with the equation “foreseeable capability = criminal intent.” That bridge is structurally unsound.