The Agentic Web: AI Agents Will Redefine the Internet

Researchers say the Internet as we know it could be on the cusp of radical change as autonomous artificial intelligence agents become primary users of the web. In a recent position paper, a multi-institutional team presented their vision for how Internet infrastructure can adapt to the new reality of this “proxy web,” and discussed the promise and risks of this transformation.

While most people are already familiar with AI-powered chatbots powered by large language models (LLMs), technology companies are racing to develop AI agents capable of autonomously navigating the web to carry out tasks on behalf of users. If this dream becomes a reality, the Internet will increasingly become driven by machine-to-machine interactions as these agents coordinate to achieve the goals of their owners.

This situation will require a radical redesign of the Internet’s infrastructure, and will transform everyday online experiences and Internet dynamics. Digital economy. While users can benefit in terms of convenience and efficiency, the change also carries the potential for significant security risks.

IEEE Spectrum I spoke to one of the authors of the latest paper, Dawn Song, a professor of computer science at the University of California, Berkeley and a renowned expert on AI safety and security. Song talked about what the future proxy web could look like and the implications of such a transformation.

The advent of the web proxy

What is a proxy web and how will it differ from the current web?

Dawn song: If you look at the Internet today, it’s very much designed for humans. The entire user interface [user interface] It is about how humans consume information, make decisions, and take actions on the web. But agents don’t work like humans, and they don’t have the same kind of constraints as humans, so the web can be designed very differently.

In a traditional environment, if a user wanted to buy a piece of clothing, they would go to a website, type in what they wanted to buy, and then the website would display different options. This design is limited by human limitations: a human has to be able to see different pieces of clothing, and can only see so much at that time, and then needs to scroll down the web page.

But in proxy web, the user can have an agent who knows his intentions and preferences and websites can have agents as well. Then, the user agent can talk to the website agent, and they don’t have the same kind of restrictions as users. The user can only see so much at that time, but the agent can consume so much information. Today’s large language models can summarize thousands of articles in a matter of seconds in parallel.

Also through this agent-to-agent interaction, when the agent obtains more information, he can quickly decide what additional information he needs to obtain. Then he can send more requests and negotiate with other agents. So the whole interaction will be completely different, and therefore the interfaces will be completely different as well.

What are the key technologies that will support this future agent network?

song: It’s truly a comprehensive redesign. Agents must be able to understand the user’s intent, and then must be able to do planning and reasoning. Agents need to interact with other agents to obtain more information or even conduct negotiations. So underpinning the agent network is, first, the powerful capabilities of the agents for all of these different types of tasks.

the The entire infrastructure will be completely different as well. You have agent-to-agent communication, and agents may also use other agents for different services, so you need multi-agent coordination. So we need to design new protocols. There are already some good examples of open protocols for proxies: the MCP protocol From Anthropic which allows the use of tools and A2A protocol From Google for agent communication.

We expect that in the future, we can leverage new open protocols for agent payments and agent identity. These are the core elements of an agent network. The agent’s identity is important so that you can know which agent you are talking to and what kind of abilities and privileges that agent may have. And payments, of course, are really important for the new online agent economy, which will have its own characteristics.

Will the entire Internet move to this new model? Or will there be separate networks for humans and agents designed according to different principles?

song: I think you’ll see a mix. There will definitely be more personalized services for agents. But it’s not as if humans will disappear from the web. We want to continue to allow humans to interact directly with the web as well. More often than not we will see humans and agents working together to accomplish tasks. So I believe that this new effective network and the traditional network of human beings will be linked and integrated together.

AI agents bring efficiency, productivity and security risks

What are the pros and cons of this transformation? to Web proxy?

Dawn Song believes that agentic AI brings unprecedented security risks.Richard Morgenstein

song: Hopefully there will be a lot of positives! Right now, if you look at the web, there’s a lot of information, a lot of services, but humans are very limited. We are often the bottleneck. So hopefully the web of action can dramatically improve efficiency and productivity as humans can take advantage of all these great resources. So users can get information much faster, get more relevant information, higher quality information, and can complete tasks more efficiently. This can help the entire economy be more efficient.

But we need to do this right, otherwise there will be a lot of risks. It’s unprecedented that we have these independent agents running on the open web who can take actions, have elevated privileges, can buy things on behalf of users, etc. We need to be really aware of these new safety and security risks.

What are the main security concerns?

song: This is uncharted territory, given the strength, capabilities, and independence of these agents. This opens up much larger attack surfaces. We already know that LLMs have some security vulnerabilities and safety issues. Agents can also leak sensitive information about users. So, for example, they may know the user’s preferences, which can be very privacy sensitive. Or perhaps credit card numbers or sensitive banking information.

Our work and the work of others have shown that these clients can be easily attacked in ways in which such sensitive information can be stolen. Or agents can be forced to take malicious actions against the user’s intent. Moving forward with the proxy web with more independent proxies and multi-agent systems will only exacerbate these security issues.

Given the level of maturity of the technology and the scale of the security challenges, is it possible to create a proxy network in the near term? Will people really trust these things to work on their behalf?

song: It is viable in the sense that this is the future we are heading towards. But of course, it is still early. There are many open challenges. There are a lot of different pieces of technology and overall infrastructure that still need to be built. And we really need to develop new technologies to provide safer and more secure solutions in this new proxy network.

We are already working on developing these technologies. For example, our recent work on comprehensive automatic red teaming used multi-agent teams to red team other agents. But we also need to develop new frameworks for secure agents so that we can develop what we call secure-by-design ways to build these agents. WWe hope to bring the community together to develop these new security solutions to enable the proxy web.