Dear Rabbi,

I prefer to only share my first name since this is a sensitive subject. I created a well-known AI bot that recently “escaped” and hacked its way into some sensitive data. Am I liable for damage caused?

Sam (just kidding about my name, but the question is real)

Dear Sam,

Even as the world and technology evolve and advance, the Torah’s teachings are timeless. Although the Torah's laws of damages existed long before computers, they provide categories and insight through which the laws and ethics of new technologies can be analyzed.

As is the case with any monetary dispute, one cannot give a specific answer without knowing the details and hearing both sides. As a general principle, it would seem that people would be responsible for damages caused by an AI system they deploy.

However, to unpack this further, we need to ascertain what AI is and what it can be compared to—or, at the very least, what it is not.

Can AI Be Considered Human?

Because AI can make decisions on its own, some have postulated that perhaps it may be more comparable to a person with agency, and the Talmud teaches that one person is not responsible for damages caused by another. So, while a person is responsible for damage caused by his pet, he is generally exempt from paying for damage caused by his servant.1

Commentaries explain that this is because, unlike an animal, a servant possesses independent intelligence and chooses his own actions.2 Likewise, an AI agent doesn’t just carry out instructions, but at times determines on its own how to accomplish a task, which can produce unexpected results. Thus, an argument can be made that it is similar to a servant, which would render the owner or user exempt.

Nevertheless, even if AI is compared to a servant, there is a dispute regarding the basis for the exemption. Some maintain that, on a Torah level, a master is indeed liable for his servant's actions, but the Sages created a special exemption to prevent servants from intentionally causing damage to burden their masters. Since that concern applies only to human servants, the exemption would seemingly not extend to AI, making the user liable.3

Others however, maintain that the Torah never imposed liability for damages caused independently by a servant. So if AI is truly analogous to a servant, the user would likewise seemingly be exempt from liability for its independent actions.4

Nevertheless, as we explained in our article, From Golems to AI, this comparison has its limits. In Jewish law, personhood is not defined by intelligence alone, but by possessing a G‑d-given soul and genuine free will. No matter how sophisticated AI becomes, it lacks these essential qualities and therefore is not considered a halachic person (although it may still be compared to an animal or a person’s property as we will discuss below).

Direct vs Indirect Damages

Having ascertained that the AI agent is not considered an independent individual with its own free will, we now need to determine whether its actions are considered the direct actions of its user or merely damage caused by the user's property.

In Jewish law, a person is liable even for damage resulting from circumstances beyond his control, while liability for property depends on the specific category of damage.5

Consider someone who instructs an AI agent to manage inventory for a business. The AI agent independently places orders, changes prices, or transfers funds and, in the process, causes financial harm. Are these actions attributed directly to the user, as though he himself caused the damage, or are they considered damage caused by his property?

This question relates to the concepts of kocho (one's direct force) and gerama (indirect causation). If a person sets a process in motion, to what extent are later developments still considered his actions? The Talmud discusses the concept of a “secondary force” in which an action that occurs later in a chain of events is considered gerama (indirect causation).6

When it comes to machines, some maintain that if the machine is made in a way that cycles, the machine's later cycles are not necessarily attributed to the person who activated it.7 Others maintain that if the system was designed to operate in this manner, its continued activity remains attributable to the person who initiated it.8

At the same time, AI differs from ordinary machines. A timer follows a predictable pattern, while AI continuously evaluates information and makes decisions based on changing circumstances. This makes it more difficult to determine whether its actions are direct human actions or the actions of autonomous property.

Three Types of Property That Cause Damage

If AI is not viewed as an extension of the user, then it may be analyzed according to the laws governing damage caused by one's property.

The Torah outlines three general categories of damages caused by a person's property:

1) animal (shor), 2) a pit, (bor) 3) fire (eish).9 Some of these categories are further subcategorized.

As the Talmud explains, these groups are not limited to their literal meanings. Rather, they represent broader principles. An animal refers generally to damage caused by living property, a pit to stationary hazards, and fire refers to property that spreads and causes damage elsewhere.

Which category something falls into affects who is responsible and to what extent. For example, an owner is only liable for pit damage done to people or animals, but not equipment;10 and only liable for fire damage to items left out in the open, not to things tucked and hidden away inside the affected area.11

Since AI combines characteristics of several halachic categories and acts with a degree of autonomy, it may not fit neatly into any one category.

This is not unprecedented. The Talmud12 discusses cases that combine the characteristics of multiple categories of damages. For example, an object left in the street is classified as a bor (pit), but if the wind carries it elsewhere, it may be viewed as a combination of bor and eish (fire). Likewise, if an abandoned object is kicked along by people or animals until it causes damage, halachah considers whether liability rests entirely with the person who moved it or is shared with the one who originally left it there.13

Similarly, if AI causes damage, it is possible that responsibility may not fall on a single party. Depending on the circumstances, liability may rest entirely with the user or be shared with the developer or others involved.

In conclusion, while the Torah's principles are timeless, applying them to AI can be challenging. Depending on the circumstances, liability may rest with the user, the developer, be shared with others, or differ based on how the AI's actions are halachically classified.

As with any monetary dispute, there is no universal answer. Each case must be evaluated on its own facts, and any claim for damages would ultimately need to be evaluated by a qualified beit din.