Guillermo Vargas

Just a few years ago, generative AI models entered the media spotlight. The dizzying pace at which these technologies evolve quickly turns yesterday’s revolution into just another item on the endless list of weekly disruptive innovations. Amid this whirlwind of technological advancement, the term AI agents is gaining increasing traction. These autonomous systems are experiencing exponential growth, prompting us to rethink their supposed autonomy, decision-making capabilities and, above all, the level of trust we can — or should — place in systems that operate without constant human supervision. As digital transformation continues to reshape how we interact with the world, it’s only natural that concerns are growing about the transparency, security and governance of these autonomous systems. Blockchain technology is the key to solving the trust problem in autonomous systems. In this context, at Telefónica Tech’s Blockchain team, we are convinced that Blockchain technology itself is the key to addressing the issue of trust. This decentralized infrastructure provides mechanisms to verify actions and decisions, preserving system integrity and striking the right balance between technological innovation and the necessary trust in independently operating systems. What are AI agents and why are they transforming the digital landscape? While the concept of intelligent agents isn’t new, it’s hard to ignore the momentum it has gained in recent months. But what exactly are AI agents? For years, we’ve developed APIs (Application Programming Interfaces) as structured languages for computer systems to communicate with one another. These interfaces have been essential to the integration of digital services but have always required human orchestration or specific programming for each interaction. AI agents mark a qualitative leap in this paradigm: they not only respond to instructions, but also maintain a persistent state, possess their own initiative, and continuously learn from interactions. If traditional APIs are the vocabulary systems use to communicate, AI agents are full-fledged interlocutors with memory, intent, and adaptability. —Example: When an AI agent schedules a meeting, it doesn’t just execute a simple command. The process involves a sophisticated sequence of actions: checking calendars, identifying participant preferences based on historical patterns, proposing viable alternatives, managing responses, and adjusting parameters as unexpected situations arise. All of this happens autonomously, without human supervision at each step. Unlike traditional reactive AI systems that respond to specific queries and then 'forget' the context, agents maintain a continuous understanding of their goals and the environment in which they operate. This persistence enables them to handle complex tasks over time, adapt to changing circumstances and learn from each interaction to enhance future performance. AI agents represent a remarkable shift in how we interact with technology. Current limitations: the challenge of trust and autonomy Despite their enormous potential, AI agents face significant limitations in their current implementation that hinder true autonomy: Centralization and dependency AI agents typically operate within centralized infrastructures. This creates single points of failure, third-party dependency, and limited interoperability. If a problem arises, the operation of all dependent agents can be disrupted. Moreover, the capabilities and rules under which these agents operate may be constrained by the technical and policy specifications of the infrastructure, limiting their flexibility and interoperability. Lack of transparency and trust An agent’s internal workings can be complex and not always transparent (often likened to a “black box”). This makes it difficult to verify the decisions these systems make. The absence of a clear, auditable log of actions and the difficulty in assigning accountability in the event of errors or issues can diminish trust in large-scale implementations. Autonomous coordination and oversight AI agents face significant challenges when attempting to carry out economic transactions autonomously. These limitations are often due to strict regulations designed to prevent fraud and malicious attacks, which restrict the ability to execute programmatic transactions freely. Additionally, the lack of effective incentive mechanisms to align the interests of multiple autonomous agents and the difficulty of collaboration without centralized supervision pose further barriers. ■ These factors together represent significant obstacles that limit the ability of agents to operate independently in complex economic environments. These challenges highlight the need to evolve toward more decentralized and transparent systems that can support true autonomy and the trust required for AI agents to deliver on their transformative promise. These autonomous systems compel us to reconsider their decision-making capabilities and the level of trust we can place in them. Blockchain as the solution: the infrastructure for truly autonomous agents From the AI agent perspective, Blockchain technology emerges as a foundational infrastructure, addressing each of these limitations head-on: Decentralization and true autonomy Blockchain provides a decentralized infrastructure where agents can operate without relying on a central authority. Its distributed nature eliminates single points of failure, allows agents to exist and operate without control from a specific entity, and establishes a common standard that facilitates interoperability across agents from different origins. Verifiable transparency and trust All agent actions and decisions are permanently and immutably recorded on the blockchain. Any party can verify an agent’s historical behavior, trace the origin of every decision and action, and achieve a level of transparency unattainable in centralized systems. Agent economy Cryptocurrencies enable frictionless micropayments between agents; smart contracts formalize and execute agreements automatically; and the economy built on Blockchain aligns the incentives of independent agents, giving rise to an entirely new economic ecosystem. ■ The combination of Blockchain and AI agents enables a system where agents can interact, collaborate, and compete autonomously, creating value and solving complex problems without constant human intervention. With Blockchain, all agent actions and decisions are recorded permanently and immutably. The future: an economy of autonomous agents The synergy between AI agents and Blockchain is essential for fostering autonomy, transparency and trust in the digital age. Although challenges remain in terms of scalability and efficiency, the convergence of these technologies promises a more decentralized, transparent and trustworthy technological future. This convergence is laying the foundation for a new digital economy where: Autonomous agents can deliver services, negotiate, collaborate, and compete. Humans can delegate complex tasks to coordinated teams of agents. The transparency and verifiability of Blockchain ensure these systems are trustworthy. Economic mechanisms align the incentives of all participants. This approach not only enhances the autonomy and capabilities of AI agents, but also increases user trust in these systems — a crucial factor for their adoption and application across countless sectors. AI & Data The truth about the 320 seconds to hack Bitcoin: a technical analysis May 27, 2025

The use of Artificial Intelligence (AI) is experiencing exponential growth, and this brings with it new issues of debate, such as the accuracy and biases of algorithms, and more specifically when we talk about generative AI. There are also growing concerns about the privacy and security of our data as this digital revolution advances and redefines our interaction with the world. This is where Zero-Knowledge Proofs (ZKP) emerge as a highly promising solution. These tests allow the verification of information while preserving the confidentiality of sensitive data, which represents an excellent balance between technological progress and the protection of individual privacy. How do Zero-Knowledge Proofs (ZKP) work? Although the premise of ZKP may seem almost magical, this technique within cryptography is not new, in fact, its origins and early academic papers date back several decades. This technology is a cryptographic tool that makes it easy to prove the veracity of a statement without the need to reveal additional information. In essence, ZKPs allow us to prove that we know something without the need to disclose how we found it out. The implementation of ZKP today could represent a revolution in any industry that adopts it. Although there are many simplified examples to explain its application, it is essential to have a clear understanding of how this technology really works. Consider a business relationship between two companies: company A has a specific need, while company B has the capacity to satisfy it. This type of business relationship is based on mutually beneficial agreements, where one party solves a problem and the other receives compensation for providing the solution. Yet, the transactions are not as simple in practice as a straightforward exchange of goods or services for payment. These arrangements often involve additional complications, such as the involvement of lawyers and the negotiation of lengthy contracts outlining the terms of the service provided. Then a relevant question arises: what happens if the company providing the service takes longer than expected to find or implement the solution? Most likely, the first company will not make payment until it is certain that the solution has been found, while the service provider may be reluctant to invest resources without guarantees of compensation. This scenario may appear to be a deadlock with no obvious solution. This challenge is addressed from the perspective of the ZKP by establishing two clear roles: The 'prover', who claims to have the solution. The 'verifier', who must confirm the correctness of this proposed solution. The 'prover' must prove to the 'verifier' that the proper solution is known, as many times as necessary, but without revealing how it was obtained; only partial evidence of the final result is presented. ✅ Although it could be argued that the 'prover' does not actually possess the solution and that his success could be due merely to luck, the possibility of repeating this process as many times as necessary statistically reduces this probability to a negligible level. An example of a Zero Knowledge Proof: The cave and the magic word A simple way to explain Zero Knowledge Testing (ZKP) is found in the 1990 article, 'How to Explain Zero Knowledge Protocols to Your Children?' Let us imagine that two individuals are in a cave divided by two paths blocked by a magic door. This door can only be opened by means of a secret word: One of the individuals, to prove that he/she knows the word without revealing it, selects one of the paths and stands in front of the magic door. The other person, from the other position, indicates which path to take to return to the magic door. If he/she manages to return correctly and consistently, no matter the path previously chosen, it is clear that he/she knows the secret word since only with it he/she could have opened the door. Source: Wikipedia. After understanding this example, we can conclude that any protocol based on Zero Knowledge Proofing (ZKP) technology must adhere to the following three fundamental properties: Soundness: An individual will only convince the other if he/she is indeed telling the truth. Completeness: If the individual is telling the truth, there is a high probability that he/she will eventually succeed in convincing the other. Zero knowledge: The observer will not learn additional information about how the problem is solved, beyond the truthfulness of the statement. While this set of requirements may seem extraordinarily promising and even unrealistic, it is supported by a series of intricate mathematical concepts that make it possible to perform the necessary calculations. ✅ These advanced mathematical techniques, including cryptography and number theory, provide a solid theoretical foundation that enables ZKPs to operate securely and reliably. ZKP in Blockchain and Machine Learning: applications and benefits In this sense, Blockchain technology plays an essential role in the development and implementation of ZKPs, as they share the common goal of ensuring privacy and strengthening trust in areas where it is challenging. The synergy between the two technologies reinforces the approach to protect privacy and address the challenges inherent in digital environments. This synergy manifests itself particularly effectively in Layer 2 Blockchains, where many use ZKP as the basis for their rollups, offering improved scalability and efficiency while maintaining the security of the base layer. This mechanism makes it possible to handle a high volume of transactions faster and at lower cost, while maintaining security and data integrity on the main chain. In addition, by using ZKP, rollups can offer an additional level of privacy, as specific transaction details do not need to be revealed on the main chain (only a cryptographic proof that proves the validity of these transactions). ✅ In the context of artificial intelligence, Zero-Knowledge Machine Learning (ZKML) integrates machine learning with zero-knowledge testing, allowing AI models to be trained on sensitive data without revealing sensitive information about that data. ZKPs have the potential to innovate and improve machine learning models in several ways: Verification: Enable verification of AI processes without exposing sensitive data, providing security and confidence in the operation of the models. Security: Protect the integrity of AI models, ensuring that they are not altered or manipulated, which is essential in insecure environments such as public clouds or edge devices. Privacy: They facilitate the training of machine learning models with private data without exposing such data to model creators or users, enabling their use in sensitive sectors such as healthcare or finance, without compromising privacy. The ability of ZKPs to test the validity of computations without revealing underlying data thereby naturally aligns with the need for transparency in AI processes, from training to execution. This alignment enables a natural fit between AI and Blockchain chains, facilitating traceability of the steps models take from data. The synergy between ZKP and AI is critical to foster privacy and trust in the digital age. The combination of these technologies, while challenges remain in terms of scalability and efficiency, promises a more secure, transparent and privacy-friendly technological future. The ZKPs, by providing a means to verify the integrity of AI processes without compromising data confidentiality, pave the way for wider and more trusted adoption of AI. This approach not only enhances the protection of sensitive data, but also increases user confidence in AI systems, crucial for their acceptance and application in various industries. AI of Things Blockchain Blockchain reinvents Digital Identity April 12, 2023