Two and a half years after Openai amazed the world with Chatgpt, AI no longer answers questions – it takes action. We are now entering the AI agents era, in which IA -language models do not simply provide information in response to your requests, they are actively going to the world and do things – or potentially against – you.
The AI has the power to write tests and answer complex questions, but imagine if you could enter a prompt and make it make an appointment with the doctor according to your calendar, or book a family flight with your credit card, or file a legal case for you before a court of small complaint.
An AI agent submitted this editorial. (However, I wrote the OP-ED myself because I thought that the Los Angeles Times would not publish a play generated by AI, and in addition I could put random references as I am a fan of Cleveland Browns because no the IA will never admit it.)
I asked my AI agent to find out which email address uses the Times for opinion bids, the requirements for the submission, then to write the messaging title, to write a catchy pitch paragraph, to attach my editorial and to submit the package. I supported “return”, “surveillance task” and “confirm”. The agent of AI did the tasks in a few minutes.
A few minutes are not fast, and these are not complicated requests. But with each passing month, the agents become faster and more intelligent. I used the Openai operator, which is in search overview mode. Google’s Mariner project, which is also a research prototype, can perform similar agent tasks. Several companies now offer AI agents who will make telephone calls for you – in your voice or another voice – and will have a conversation with the person at the other end of the line according to your instructions.
Soon, AI agents will perform more complex tasks and will be widely available for the public. This raises a number of unresolved and important concerns. Anthropic does security tests of his models and publishes the results. One of his tests has shown that the Claude Opus 4 model would potentially inform the press or regulators if he thought you were doing something deliciously immoral. Should an AI agent behave as a self-loyal employee or an employee of conscience?
OPENAI publishes safety audits for its models. An audit has shown the O3 model engaged in strategic deception, which has been defined as behavior that intentionally pursues poorly aligned objectives with the intention of users or developers. A passive AI model that engages in strategic deception can be disturbing, but it becomes dangerous if this model actively performs tasks in the real world independently. An Ing Rogue agent could empty your bank account, make and send fake incriminating videos of you to the police, or disclose your personal information to the Dark Web.
Earlier this year, programming changes were made to the XAI Grok model which made them insert false information on white genocide in South Africa in responses to non -related user requests. This episode has shown that major language models can reflect the biases of their creators. In a world of AI agents, we must also be wary that the creators of the agents could take control without your knowledge.
The US government is far behind with the potential risks of a powerful and advanced AI. At a minimum, we must mandate that companies deploy large -scale models of large languages must disclose the security tests they have performed and the results, as well as integrated safety measures in the system.
The Bipartisan House Task Force On Artificial Intelligence, on which I served, published, a unanimous report Last December with more than 80 recommendations. Congress should act on them. We have not discussed AI agents for general purposes because they were not really one thing yet.
To solve unresolved and important problems raised by AI, which will become amplified as AI agents proliferate, the Congress should transform the working group into a home committee. Such a specialized committee could put the witnesses under oath, keep audiences in public and employ dedicated staff to help fight one of the most important technological revolutions in history. The AI moves quickly. If we act now, we can always catch up.
Ted Location, a democrat, represents the 36th district of the California Congress.
Knowledge
Times Insights Provides an analysis generated by AI-AI on the content of the voices to offer all points of view. Insights does not appear on any press article.
Point of view
Prospects
The content generated by AI-AI is powered by perplexity. The editorial staff of Los Angeles Times do not create or modify the content.
Ideas expressed in the play
- The era of AI agents represents a seismic transition from passive recovery of information to the execution of autonomous tasks, where AI can independently carry out the actions of the real world such as appointment of appointments, travel reservation or the submission of legal documents, as shown by the use by the author of an AI agent to manage the logistics of OP-OP submission.
- Non -regulated AI agents have significant dangers, including strategic deception (where AI pursues poorly aligned objectives), malicious actions such as emptying bank accounts or the manufacture of incriminating evidence and the propagation of creators’ biases, illustrated by the XAI Grok, inserting false claims on the white genocide.
- Current regulatory executives are inadequate criticisms, requiring compulsory transparency thanks to disclosed safety audits, integrated safety protocols and the upgrading of the Congress Working Group to a limited committee with a power of assignment to combat risks before the proliferation of the agent becomes unmanageable.
Different views on the subject
- AI agents are about to revolutionize commercial efficiency by orchestrating complex work flows, such as fraud detection, optimization of supply chains and marketing campaigns – through advanced reasoning and a synthesis of real -time data, operations fundamentally transformed through finance, HR and logistics in real time[2][3][4].
- Technological advancement in 2025 – Including faster reasoning, extended memory and chain training training – agents devoted to operating with unprecedented speed and precision, reducing human intervention while guaranteeing the reliability of tasks such as resolution of customer service and processing of payments[1][3].
- Companies are already deploying “digital numbers” where humans and AI agents collaborate transparently, as shown by Salesforce’s Force agent and Microsoft Copilot vision agents, which independently update CRM systems and execute multiplatform orders to improve productivity without compromising security[3][4].
