The tech sector follows specific regulations, just as the rest of the world.
As personal computers became more popular, USB emerged as a standard for data transfer between devices. The internet gave rise to IP addresses, which are numerical labels that are used to identify all devices on the internet. Email was introduced together with SMTP, a foundation for sending emails over the internet.
As technology advances, new protocols are created to regulate how things interact, communicate, and function. These protocols are the unseen framework of the digital world.
The world must create new ones as it enters an AI-shaped era. But artificial intelligence is more than just screens and code. It compels programmers to reconsider basic issues about the interplay between online and offline technical systems.
How will AI and humans coexist? In what ways will AI systems interact with one another? How are we going to design the protocols that will govern a new era of intelligent systems?
Both big giants and startups in the sector are working hard to create protocols that will address these problems. AI models are still primarily controlled by humans in the present. There are some who are preparing for a future where a large portion of human labor has been replaced by AI.
According to Antoni Gmitruk, chief technology officer of Golf, a company that assists clients in deploying remote servers that are in line with Anthropic’s Model Context Protocol, protocols will be this type of standardized method of processing non-deterministic data. The nature of agents and artificial intelligence in general is “inherently non-deterministic in terms of what they do and how they behave.”
The easiest way to deal with unpredictable AI behavior is to envision potential outcomes and test them using fictitious scenarios.
Some of them require explicit protocols
Scenario 1: AI and humans conversing on an equal footing
One technique to find protocols that achieve the ideal balance of power between humans and AI is through games.
A group of youthful professionals in cryptography introduced Freysa, an AI agent that allows human users to manipulate it, at the end of 2024. Unlike other rules, the prize is yours if you can make Freysa fall in love with you or agree to give up its funds. When human intuition and computer logic are at odds, the reward pool increases with each unsuccessful effort.
Prominent figures in the tech sector have taken notice of Freysa, including seasoned venture capitalist Marc Andreessen and Elon Musk, who described one of its games as “interesting.”
One of Freysa’s architects told in a January interview, speaking on condition of anonymity, “The main technical thing we’ve done is enabled her to have her own private keys inside a trusted enclave.”
In the tech sector, secure enclaves are nothing new. Businesses ranging from Microsoft to AWS utilize them as an additional security measure to separate sensitive information.
According to the architect, Freysa’s situation is the first step toward becoming a “sovereign agent.” He described that as the kind of agent that will probably proliferate: one that has the ability to manage its own private keys, access funds, and change on its own.
And why are we doing it now? According to the architect, we’re about to enter a stage where AI is becoming sufficiently advanced to let you envision the future, in which it will essentially replace all of our labor, including yours and mine, and become economically productive as independent entities.
They said that during this stage, Freysa contributes to the resolution of a fundamental query: What constitutes human involvement? And how is human co-governance over agents at scale achieved?
Eternis AI is the startup behind Freysa, according to a May report from the cryptocurrency news website The Block. Eternis AI identifies itself as an applied AI lab that focuses on sovereign agent systems, multi-agent coordination, and enabling digital twins for everyone. Coinbase Ventures is among the investors who have contributed $30 million to the startup. Augustinas Malinauskas, Ken Li, Pratyush Ranjan Tiwari, and Srikar Varadaraj are its co-founders.
Scenario 2: To the current architects of intelligence
Freysa creates procedures for a fictitious future in which AI agents and humans engage with comparable degrees of autonomy. However, as AI is still a creation of human design and intent, the world must also establish guidelines for the present.
According to Davi Ottenheimer, owner of security consultancy Flyingpenguin and a cybersecurity expert who focuses on the nexus of technology, ethics, and human behavior, AI usually operates on the web and builds upon pre-existing protocols. He stated, “But it adds in this new element of intelligence, which is reasoning,” yet procedure for reasoning is still lacking.
All of the news seems to be hinting at this. Yes, they scanned all of the books ever written without ever asking permission. “Well,” he responded, “there was no protocol that said you can’t scan that.”
Although there are laws, there may not be protocols.
The Authors Guild is suing OpenAI for copyright after the company used data from “more than 100,000 published books” to train its models before erasing the data. Meta contemplated purchasing Simon & Schuster outright in order to obtain published literature. Moreover, in order to train their AI models, IT companies have turned to using nearly all of the consumer data that is accessible online, including the remnants of social media sites like Friendster and Myspace as well as the content of open Google Docs.
Ottenheimer compared the present data dash to the development of ImageNet, the visual database that drove computer vision, which was constructed by Mechanical Turk employees who searched the internet for content.
“They did a bunch of stuff that a protocol would have eliminated,” he explained.
Scenario 3: Getting along with one another
As artificial general intelligence becomes a reality, protocols for intelligent systems’ communication with one another and the outside environment will be necessary. These systems could range from foundation models to agents.
In order to set the path, the top AI businesses have already introduced new ones. In November 2024, Anthropic, the company that created Claude, introduced the Model Context Protocol, or MCP. According to this description, it is a universal, open standard that replaces fragmented integrations with a single protocol to link AI systems with data sources.
A protocol called Agent2Agent was introduced by Google in April. It will enable AI agents to coordinate operations, securely exchange information, and communicate with one another on top of different enterprise platforms or apps.
In addition to addressing new interoperability and scaling issues that have proven crucial to AI adoption, these expand upon current AI protocols.
Therefore, he continued, controlling agent behavior is a first step before releasing the full potential of AGI and allowing them to roam the globe unhindered. When that time comes, Gmitruk stated, agents will speak in normal language instead of using APIs. They will need to be verified and will have distinct identities and even occupations.
How can we allow agents to talk with one another? They shouldn’t just be computer programs running on a server; they should be some type of existent entity with a history and goals of their own, according to Gmitruk.
Setting guidelines for agent-to-agent contact is still too early, according to Gmitruk. He started a startup earlier this year with the goal of developing an agent authentication system, but he and his colleagues changed their minds.
He stated on LinkedIn that it was too early for agent-to-agent authentication. Although they still believe that agent-native access to the traditional internet is necessary, they have stepped up their support for MCP because it is more pertinent given the current state of agents.
Does every situation require a protocol?
Certainly not. The AI boom represents a turning point, reviving discussions about the distribution and commercialization of knowledge.
It is a “inflection point” in the fourth industrial revolution, according to McKinsey & Company. This wave of change started in the middle of the 2010s and includes the current period of “connectivity, advanced analytics, automation, and advanced-manufacturing technology.”
Such incidents bring up an important question: To what extent does the market own innovation and to what extent does the public? The argument about the merits of closed and open-source models in the field of artificial intelligence is the clearest example of this.
“I think we will see a lot of new protocols in the age of AI,” said Tiago Sada, chief product officer at Tools for Humanity, the firm that developed the technology used in Sam Altman’s World story. nevertheless, “I don’t think everything should be a protocol.”
World is a system created for a day when people will always need to prove who they are. Any protocol, according to Sada, “should be like this open thing, like this open infrastructure that anyone can use,” and be unaffected by influence or restriction.
However, he noted that one drawback of protocols is that they can occasionally be slower to proceed. When was the last time a new feature was added to email? or the web? Although protocols are inclusive and open, he noted that they can be more difficult to innovate and monetize. Therefore, while some things in AI will be developed as protocols, many will still remain products.
