These protocols will help AI agents navigate our messy lives

What should these protocols say about security?

Researchers and developers still do not really understand how artificial intelligence models work, and new weaknesses are discovered all the time. For AI applications similar to Chatbot, malicious attacks can cause all kinds of bad things, including renewing training and waste data. But for artificial intelligence agents, who interact with the world on behalf of someone, the possibilities are more dangerous.

For example, it has already turned out that one of the artificial intelligence agents, who was conducted on reading and sending email messages to a person, is subject to what is known as the indirect injection attack. Basically, an email can be written in a way that kidnaps the artificial intelligence model and leads to a defect. Then, if this agent has access to user files, it may be directed to send private documents to the attacker.

Some researchers believe that protocols such as MCP should prevent agents from carrying out harmful actions such as this. However, it is not at the present time. “Basically, she has no security design,” says Zhaor Chen, a doctorate student at the University of Chicago and works on the security of AI’s agent and uses MCP servers.

Bruce Schinier, a researcher and security activist, is skeptical that protocols like MCP will be able to do a lot to reduce the inherent risks that come with artificial intelligence and that giving this technology more power will give it more ability to cause harm in the real material world. “We have no good answers about how to secure these things,” says Shen. “Cisbul will be truly fast.”

Others are more hopeful. Safety design can be added to the MCP and A2A similar to the way in which the Internet protocols such as HTTPS (although the nature of attacks on artificial intelligence systems is completely different). Chen and Anthropor believe that monotheistic protocols such as MCP and A2A can help facilitate safety problems and solve them even as it is. Chen MCP is used in his research to test the roles that various programs can play in attacks to better understand weaknesses. Chu At HotHROPIC believes that these tools can allow cybersecurity companies more easily with attacks against agents, because it will be easier to empty someone who sent what.

How open these protocols?

Although the MCP and A2A are among the most popular and popular agent protocols today, there are many other protocols in business. Large companies such as CISCO and IBM are working on their own protocols, and other groups have developed different designs such as Agora, which are designed by researchers at Oxford University, which upgrade the agent’s service connection from the human language to organized data in the actual time.

Many developers hope that there will be a record of safe and reliable systems to move in the spread of agents and tools. Others, including Chen, want users to be able to assess different services in something like Yelp for artificial intelligence agent tools. Some specialized protocols have built a block on the head of MCP and A2A so that the servers can show that they are not just unwanted messages.