AI is managed by protocol No one agrees yet

The tech industry, like everything else in the world, adheres to certain rules.

With the boom in personal computing, USB, the standard for transferring data between devices, has emerged. With the rise of the Internet, numerical labels and IP addresses have emerged that identify all devices online. With the advent of email, SMTP, a framework for routing emails across the Internet, has come.

These are protocols (an invisible scaffolding in the digital realm), and with all technological changes, new things emerge to govern how things communicate, interact and work.

As the world enters an era shaped by AI, we need to create something new. However, AI exceeds the normal parameters of the screen and code. Developers force them to rethink basic questions about how technological systems interact across virtual and physical worlds.

How do humans and AI coexist? How do AI systems interact with each other? And how do you define protocols that manage a new era of intelligent systems?

The industry-wide, startups and tech giants are busy developing protocols to answer these questions. Some dominate the present day, where humans still primarily control AI models. Others are building for a future in which AI has taken over a significant portion of human labor.

“The protocol will become this kind of standardized way of processing non-deterministic information,” Golf Chief Technology Officer Antoni Gumitrouk told BI that it will help deploy remote servers tailored to humanity's model context protocols. Agents, and in general, “it's inherently indeterminate in terms of what they do and how they behave.”

When AI behavior is difficult to predict, the best response is to imagine possibilities and test them through hypothetical scenarios.

Below are some that require a clear protocol:

Scenario 1: Humans, AI, and equality dialogue

Games are one way to determine which protocols align the right power balance between AI and humans.

In late 2024, a group of young cryptography experts launched Freysa. Freysa is an AI agent that invites human users to interact with them. The rules are unconventional. You agree to make Freisa fall in love with you or to acknowledge the funds, and the award is yours. Prize pools grow with each failed attempt at a standoff between human intuition and mechanical logic.

Freysa has attracted the attention of big names in the tech industry, from Elon Musk, who called one of the games “interesting” to veteran venture capitalist Marc Andreessen.

“The central technical thing we did is that she can now have her own private key within a trusted enclave,” said one of the Freysa architects, who spoke anonymously to BI in an interview in January.

Safe enclaves are nothing new in the tech industry. They are used by companies from AWS to Microsoft as an additional layer of security for separating sensitive data.

In the case of Freysa, the architect said it represents the first step in creating a “sovereign agent.” He defined it as an agent that can control its own private key, access money, and evolve autonomously.

“Why are we doing that at this point? We're in a stage where AI is getting better enough to allow you to see the future. This basically replaces your job, my job, all your work and becomes economically productive as an autonomous entity,” the architect said.

At this stage they said Freysa will help answer the core questions: “What does human involvement look like? And how do you have joint governance over agents on a large scale?”

In May, Crypto's news site The Block revealed that the company behind Freysa is Eternis Ai. Eternis AI describes itself as “Application AI lab focused on enabling digital twins, multi-agent coordination, and sovereign agent systems for all.” The company raised $30 million from investors, including Coinbase Ventures. Its co-founders are Srikar Varadaraj, Pratyush Ranjan Tiwari, Ken Li and Augustinas Malinauskas.

Scenario 2: Towards the current intellectual architect

Freysa establishes a protocol in anticipation of the hypothetical future when human and AI agents interact with similar levels of autonomy. But the world also needs to set current rules where AI is still a product of human design and intention.

AI usually runs on the web and builds on existing protocols developed before, explained cybersecurity strategist Davi Ottenheimer Those studying the intersection of technology, ethics and human behavior are Inudd's Vice President of Trust and Digital Ethics. “But it's been added to this new element of intelligence, and that's the reasoning,” he said, and we still don't have a protocol for reasoning.

“I see it being suggested in this kind of news. Ah, they scanned every book they've ever written and weren't asked if they could do it, he said.

There may be no protocols, but there are laws.

Openai faces copyright lawsuits from the Authors Guild to train a model on data from “over 100,000 published books” and to delete the dataset. Meta considered purchasing publishers Simon & Schuster to access published books. The Tech Giants also relied on tapping on almost all of the consumer data available online from public Google Docs content and artifacts from social media sites such as MySpace and Friendster.

But in the AI era, relying solely on legal precedents is “like trying to regulate road safety without license plates, traffic lights, lane marks, or speed limits,” Ottenheimer told BI in an email.

This is a push and pull issue. The protocol is drawn to prevent it and the law is awarded.

“Protocols and legal precedents do not conflict as they are different tools for different uses,” Ottenheimer said. “The legal system is inherently reactive and you have to look at the past all the time. They wait for harm, assign liability, try to remedy. Lawyers are afraid to go all the way to the future for a day.

Scenario 3: How to tackle each other

As artificial general information approaches a real future, protocols are needed for how intelligent systems, from basic models to agents, communicate with each other.

Large AI companies have already launched new companies. Humanity, the maker of Claude, launched the Model Context Protocol, or MCP, in November 2024. It describes it as “replacing a universal, open standard, fragmented integration, for connecting AI systems to data sources with a single protocol.”

In April, Google launched Agent2Agent. This has launched a protocol that “enables AI agents to communicate with each other, exchange information securely, and adjust actions on top of various enterprise platforms or applications.”

Although they are based on existing AI protocols, they address new scaling and interoperability challenges that have become important for AI adoption.

Therefore, managing agents' actions is “a middle step before unlocking the full power of AGI and running the world freely,” he said. When we arrived at that point, Agents no longer communicate in natural language rather than via APIs. They have a unique identity, even a job, and need to be verified.

“How do you make it possible for an agent to be not only a computer program running somewhere on the server, but also to be some kind of existing entity with that history and an existing entity with a goal,” says Gmitruk.

It's too early to set standards for agent-to-agent communication, Gmitruk said. Earlier this year, he and his team first launched a company focused on building authentication protocols for agents, but they pivoted.

“It was premature to authenticate agents from agents,” he told BI via LinkedIn. “Our overall vision is still the same –> We need agent-native access to the traditional internet, but this is more relevant at the agent stage we are in, so we doubled the MCP.”

Does everything need a protocol?

It's definitely not. The AI boom marks a turning point and revives debate about how knowledge is shared and monetized.

McKinsey & Company calls it the “inflection point” of the 4th Industrial Revolution. This is a wave of change that began in the mid-2010s and spans the current era of “connectivity, advanced analytics, automation, advanced manufacturing techniques.”

Such moments raise important questions: how many innovations belong to the general public and how many markets do they belong to? It's not as clear as the AI world debate between open source and the value of closed models.

“I think we'll see a lot of new protocols in the age of AI,” says Tiango Sada from Tools for Humanity, building the technology behind Sam Altman's world. But “I don't think everything should be a protocol.”

World is a protocol designed for the future where humans need to verify their identity each time. Sada said the goal of any protocol should be “like this open one, like this open infrastructure that anyone can use,” and has not been censored or affected.

At the same time, “One of the drawbacks of protocols is that they can move slowly,” he said. “When was the last time email gained new features? Or the Internet? The protocol is open and comprehensive, but monetization and innovation can be challenging,” he said. “So, with AI, yes – some things are built as protocols, but many are still products.”

Source link