Who’s to Blame When AI Agents Screw Up?

In the past Year, Veteran Software Engineer Jay Prakash Takur Nights and Prototype of the Weekend AI agents This can make the meal and engineering mobile applications almost completely self-owned in the near future. Although his agents are surprisingly skillful, new legal questions that are waiting for the companies trying to capitalize the hottest new technology of Silicon Valley.

Agents are AI programs It can mainly move independently, to automate the tasks such as companies respond to customer questions or respond to invoices. SHATGPT and similar conversations can prepare emails or analyze the bills when required, Microsoft and other technological giants expect those agents to solve More complex functionsMost importantly, do it With little human control.

Most of the Tech industry ambitious plans To attract many agent systems, the team to change with them agency with agents All workforces. Benefits for companies are clear: saving time and labor costs. Now, the demand for technology is growing. Technical Market Researcher Gartner Approximate This agent will solve 80 percent of total customer service surveys until AI, 2029. Fiverr is a service that businesses can order freely codingers, report In recent months, the search for “AI Agent” increased by 18.347 percent.

Thakur, who is mainly taught in California, wanted to be on the front front of the area. His day work in Microsoft is not related to agents, but he was interested in AutogenicOpen Source Program for Microsoft’s building agents, since working in Amazon in 2024. Thakur says it is preparing multi-hundred prototypes using autogenous with a programming tutor. Last week, Amazon spread a similar agent developed vehicle called ropes; Google The agent suggests that they call the Development Kit.

Because the agents are expected to act autonomous, and when the mistakes are liable to the material damage. A single, large system may be controversial in accusing agents from different companies. Compared the problem of reviewing incorrect records for reconstruction of a conversation based on the notes of different people. “It is impossible to make responsibility,” he said.

Main legal advice in Openai, Joseph Fireman, San Francisco in San Francisco, which houses the Media Law Resource Center in San Francisco, tends to follow those with the deepest pockets. This means that companies are even responsible, even if the agents can create confusion with a vehicle together with an agent. (If this person was guilty, it would probably be a valuable target money, the thought goes). “I don’t think that nobody hopes to move to the consumer sitting in the cells of their mothers on the computer.” Insurance industry began to spread the scope In order to help companies, AI covers the costs of discussion for challenges.

Onion rings

Thakur practices have brought it together in the systems that require as few people as possible. A project followed replaced companions with two agents. One was trained to seek the specialized means needed to apply, and the other summarized its own rules. In the future, the third agent can use the identified tools and can follow generalized policies to make a completely new application.

When you put Thakur prototype, a search agent, according to the website, “Supports Unlimited Survey for Enterprise Users” (high-salary clients can count as long as they want). However, in deleting basic information, the generalization agent “reduced an important qualification for enterprise users.” As a user of an enterprise, such as an enterprise user, the coding agent said that a program to write unlimited surveys for foreign service. Because this was a test, no harm. If the actual life occurred, the cut management can lead to an all-unexpected system.