Background & Biography
Illia Polosukhin’s professional background combines machine learning and software development. Kharkiv Polytechnic Institute records his studies in its computer mathematics and data analysis department. He subsequently worked as a researcher and engineer at Google, focusing on language-related machine learning. His academic record is relevant because his later AI advocacy comes with a specific, documented research contribution rather than only a commercial interest in the field.
That contribution is most clearly established by the 2017 paper “Attention Is All You Need.” Polosukhin is one of its eight co-authors, and the paper identifies the authors as equal contributors. Its contribution note credits Ashish Vaswani, working with Polosukhin, with designing and implementing the first Transformer models. It separately credits other researchers for the initial self-attention direction, architectural ideas, experiments and software, providing a more precise account than a claim that one person invented modern AI.
NEAR’s own history distinguishes the next two stages of his career. Polosukhin and Alexander Skidanov began an AI venture in 2017, initially seeking to develop models that could write software from natural-language instructions. They turned toward blockchain infrastructure and founded NEAR Protocol in 2018. The resulting project addressed problems of scalability and usability that they encountered while trying to build and coordinate software across borders.
Polosukhin was appointed CEO of NEAR Foundation in 2023. He identified himself in that role in his August 2025 newsletter, and a 2026 Funding the Commons program continues to list the position. The Foundation’s current website also names him on its governing council, while NEAR AI identifies him as founder. Switzerland is the institutional anchor used in this profile because the Foundation is based there; the United States is another relevant professional market through NEAR AI’s Delaware corporate entity. Neither designation asserts his nationality or current residence.
Notable Contributions
The Transformer paper remains an important part of Polosukhin’s influence outside cryptocurrency. It presented a sequence-modeling architecture built around attention mechanisms rather than the recurrent or convolutional structures used in the systems it compared. The research tested the approach on machine translation and other language tasks. Its significance should be credited to the full collaboration, while Polosukhin’s documented implementation work provides a concrete basis for discussing his expertise.
NEAR represents a different kind of contribution: turning research and distributed-systems experience into a platform for other developers. Its architecture uses sharding to divide work across portions of the network, with the aim of expanding capacity without requiring every participant to process everything. Building and maintaining that system requires many engineers and independent operators. The current ecosystem roadmap identifies NEAR One as the core protocol research and development team, so Polosukhin’s founder role should not absorb their technical credit.
Chain abstraction gives the broader strategy a practical form. NEAR’s Chain Signatures allow accounts, including smart contracts, to request signatures for transactions on other blockchains through a multi-party computation service. This can simplify cross-network account management, but it does not make the different networks identical or eliminate the need to understand their transaction rules. The documentation also distinguishes signing outbound transactions from obtaining verified information about external chain state.
NEAR Intents takes a related approach to execution. A user or software agent describes a desired result, such as a token swap, and third-party solvers compete to provide a quote. After approval, a verifier contract on NEAR checks and settles the transaction. The significance is the attempt to hide unnecessary operational steps from users while retaining a defined authorization process; execution quality and the dependencies behind an apparently simple interface still matter.
Polosukhin has also promoted the idea of user-owned AI, arguing that governance, payments and cryptographic verification can help align intelligent systems with the people using them. WIRED’s 2024 reporting independently documented that position. His own newsletter develops it as a public thesis about incentives and control. It is a direction for research and company-building, not evidence that decentralized architecture automatically produces safer models or that every participating service is already user-governed.
The current NEAR AI product work gives that thesis more specific tests. In July 2026, the company released IronClaw 1.0, describing an agent architecture that separates decisions from actions through a common approval and coordination layer, with persistent state across interruptions. These are relevant mechanisms for software that can act on someone’s behalf. The company’s announcement establishes the release and its claimed design; independent performance and security assessment remain separate questions.
Distribution also changes the meaning of privacy claims. NEAR AI announced availability as an OpenRouter provider on September 28, 2026, but explicitly said that the routed service was not confidential inference because requests pass through infrastructure outside its confidential environment. Its direct API is the route it identifies for hardware-attested confidential computation. That distinction is consequential: the same provider name can appear in services with materially different data paths and assurances.
Why Katharos.ai Follows Illia Polosukhin
Katharos.ai follows Polosukhin because his work puts AI and blockchain claims into contact with operational systems. An agent that can transfer assets needs more than a capable model: it needs permissions, trustworthy execution, recoverable state and a way to verify what happened. Blockchain infrastructure may supply some of those components, while confidential computing may address others. His projects offer evidence for examining where those combinations help and where they introduce new dependencies.
His influence also spans institutions with different purposes. A nonprofit foundation supports an ecosystem, a protocol team develops shared infrastructure and an AI company offers products to customers. Decisions made in one setting do not automatically represent all the others. Katharos.ai pays attention to how responsibilities, funding and control are communicated, especially when a unified public narrative brings several organizations together under the NEAR name.
Areas to Follow
- Agent permissions: how systems constrain actions, protect credentials, preserve useful state and let users review or revoke authority. - Confidential computing: which requests are actually covered by attestation, who verifies it and where data leaves a protected environment. - Cross-network execution: solver competition, quote quality, bridge dependencies and the reliability of withdrawals and settlement. - Protocol development: deployed sharding improvements, operator requirements and the division of responsibility among contributor teams. - Adoption: repeat use, paying customers and application retention, with account or transaction counts kept distinct from unique human users. - Governance: public decision processes, Foundation responsibilities and evidence that proposed changes have actually been implemented.
Reporting Scope
Coverage concerns Polosukhin’s public research, professional leadership, technical explanations and the organizations developing the NEAR ecosystem. Research papers anchor scientific credit, while current documentation and release announcements establish the scope of implemented products. Personal views are identified as his views rather than treated as decisions already adopted by a foundation, protocol or company. Community discussion alone does not establish that a proposed policy has taken effect.
Privacy and safety language receives particular scrutiny. A hardware attestation concerns a defined computing environment; it is not a universal guarantee about model behavior, external tools or every service used along the way. Company benchmarks and adoption claims require their own methods and context. Katharos.ai profiles explain professional influence and do not constitute endorsements or investment advice. This profile was checked on October 6, 2026.
Areas of Influence
Exchanges & Infrastructure is the primary category for Polosukhin’s work on blockchain execution, interoperability and services supporting applications. Investment & Founders recognizes his role in establishing NEAR’s ventures and shaping their direction. Community, Media & Education reflects published research, public technical explanation and discussion of AI ownership. The connection among these areas is his continuing effort to make complex systems usable while asking who controls the infrastructure, the data and the authority to act.
Katharos.ai publishes this profile for information, based on the sources listed. Inclusion is not an endorsement of the individual, their organizations, or any investment product.
Source notice: Information and images are attributed to the external sources listed. Katharos.ai does not endorse the subjects or claims described and does not independently certify the authenticity of every source statement or image.