【Patent Division Application】US administration reportedly considering pre-screening system for AI models — Existing patent divided, aiming for multi-faceted AI model verification by AIs operated by different entities. Establishing 'Claude Mythos-grade AI' pre-screening foundation, resolving the oracle problem.
Cycale Trust announced a patent division application for its core 'Appraisal Certification System (R)' technology. This move aims to broaden the scope of rights for foundational technology that enables multiple AIs, operated by different entities, to verify the authenticity (trust) of other AIs and all information generated by them, responding to reports of the US administration considering AI model pre-screening.
📋 Article Processing Timeline
- 📰 Published: May 6, 2026 at 02:50
- 🔍 Collected: May 5, 2026 at 18:01
- 🤖 AI Analyzed: May 5, 2026 at 18:13 (11 min after Collected)
Cycale Trust Inc. (Headquarters: Shibuya-ku, Tokyo; Representative Director: Tsuyoshi Sue; hereinafter 'Cycale Trust') announced that it has filed a 'division application' for existing patents constituting its core technology, the 'Appraisal Certification System (R)'. This 'division application' aims to carve out a broader scope of rights for foundational technology that enables multiple AIs, operated by different entities, to conduct multi-faceted verification of the authenticity (trust) of not only other AIs but also all information generated by those AIs.
Chapter 1: AI Transitioning from 'Tool' to 'Entity Influencing Social Infrastructure'
Anthropic's 'Claude Mythos Preview', announced in April 2026, demonstrated its ability to discover numerous 'zero-day vulnerabilities' in critical software, including major operating systems and web browsers. Due to exploitation risks, the company withheld public release of the model, offering it as an invite-only research preview limited to defensive cybersecurity applications.
This fact indicates that AI has moved beyond merely being a tool for operational efficiency and has entered a stage where it directly impacts the very social infrastructure, such as semiconductors, finance, critical infrastructure, defense, medical care, and software supply chains. Anthropic itself has disclosed ongoing discussions with US government officials regarding the model's offensive and defensive capabilities.
Moving forward, it is anticipated that a framework resembling 'pre-screening'—including capability assessment, use-case review, and output verification prior to utilization—will be progressively strengthened in various countries for AI utilized in sectors like semiconductors, finance, critical infrastructure, defense, and medical care.
【Meaning of Technical Terms】
* 'Zero-day vulnerability': A flaw in software that has not yet been publicly disclosed and for which no corrective patch (fix program) has been provided. If discovered first by attackers, systems remain defenseless against attacks.
* 'Pre-screening': Refers to an independent procedure to verify whether an AI's capabilities, intended use, and output are trustworthy and authentic before adopting its output for business operations or policy decisions.
Chapter 2: Core Challenge: The 'Oracle Problem' in the AI Era
Blockchain possesses extremely strong resistance to tampering with information once recorded. However, the fundamental question of whether the information itself is correct *before* being recorded on the blockchain, who verified it and on what basis, or whether business systems can adopt information generated by an AI agent (agent AI)—this 'authenticity (trust) before recording to the blockchain (or initiating a blockchain smart contract)' remains unresolved. Cycale Trust views this as the 'Oracle Problem in the AI Era'. If this problem is not solved, concepts like 'Next-generation AI On-chain Initiatives' will become mere formalities.
In essence, this challenge raises the most difficult questions in AI governance: not 'Which AI is superior?', but rather 'Who, on what basis, and how verified the output of that AI?', 'Can the verifying AI itself be trusted and held accountable?', and 'Can the verification results be later audited and explained (accountability)?'
If a single large AI merely judges the safety and authenticity (trust) of another AI, it cannot resolve structural issues such as bias in the verification entity, bias in training data, conflicts of interest of the operating entity, and the black-box nature of AI models (the problem of internal workings being opaque). The mechanism of having one AI verify another AI itself carries the risk of falling into a closed verification structure.
【Meaning of Technical Terms】
* 'Oracle Problem': The fundamental problem of how to correctly incorporate real-world information external to the blockchain into the blockchain. Even with high resistance to record tampering, the veracity at the point of input cannot be guaranteed.
* 'AI Oracle Problem': The state where the 'Oracle Problem' is extended to information generated and verified by AI. It means that even if AI output is directly recorded on a blockchain, its correctness still requires separate verification.
* 'Black-box nature': Refers to a state where the decision-making process within an AI model cannot be observed or explained externally, making it impossible to trace how a particular conclusion was reached.
Chapter 3: Positioning of this 'Division Application'
The existing patent relates to the 'Appraisal Certification System (R)', a patent for evaluating verification information to guarantee the authenticity (trust) of all assets (including 'physical assets' like goods/products, 'non-physical assets' like data/content, and 'hybrid assets' like RWAs and digital twins that integrate both) using an evaluation model generated by machine learning, and outputting the verification results.
This division application...
Chapter 1: AI Transitioning from 'Tool' to 'Entity Influencing Social Infrastructure'
Anthropic's 'Claude Mythos Preview', announced in April 2026, demonstrated its ability to discover numerous 'zero-day vulnerabilities' in critical software, including major operating systems and web browsers. Due to exploitation risks, the company withheld public release of the model, offering it as an invite-only research preview limited to defensive cybersecurity applications.
This fact indicates that AI has moved beyond merely being a tool for operational efficiency and has entered a stage where it directly impacts the very social infrastructure, such as semiconductors, finance, critical infrastructure, defense, medical care, and software supply chains. Anthropic itself has disclosed ongoing discussions with US government officials regarding the model's offensive and defensive capabilities.
Moving forward, it is anticipated that a framework resembling 'pre-screening'—including capability assessment, use-case review, and output verification prior to utilization—will be progressively strengthened in various countries for AI utilized in sectors like semiconductors, finance, critical infrastructure, defense, and medical care.
【Meaning of Technical Terms】
* 'Zero-day vulnerability': A flaw in software that has not yet been publicly disclosed and for which no corrective patch (fix program) has been provided. If discovered first by attackers, systems remain defenseless against attacks.
* 'Pre-screening': Refers to an independent procedure to verify whether an AI's capabilities, intended use, and output are trustworthy and authentic before adopting its output for business operations or policy decisions.
Chapter 2: Core Challenge: The 'Oracle Problem' in the AI Era
Blockchain possesses extremely strong resistance to tampering with information once recorded. However, the fundamental question of whether the information itself is correct *before* being recorded on the blockchain, who verified it and on what basis, or whether business systems can adopt information generated by an AI agent (agent AI)—this 'authenticity (trust) before recording to the blockchain (or initiating a blockchain smart contract)' remains unresolved. Cycale Trust views this as the 'Oracle Problem in the AI Era'. If this problem is not solved, concepts like 'Next-generation AI On-chain Initiatives' will become mere formalities.
In essence, this challenge raises the most difficult questions in AI governance: not 'Which AI is superior?', but rather 'Who, on what basis, and how verified the output of that AI?', 'Can the verifying AI itself be trusted and held accountable?', and 'Can the verification results be later audited and explained (accountability)?'
If a single large AI merely judges the safety and authenticity (trust) of another AI, it cannot resolve structural issues such as bias in the verification entity, bias in training data, conflicts of interest of the operating entity, and the black-box nature of AI models (the problem of internal workings being opaque). The mechanism of having one AI verify another AI itself carries the risk of falling into a closed verification structure.
【Meaning of Technical Terms】
* 'Oracle Problem': The fundamental problem of how to correctly incorporate real-world information external to the blockchain into the blockchain. Even with high resistance to record tampering, the veracity at the point of input cannot be guaranteed.
* 'AI Oracle Problem': The state where the 'Oracle Problem' is extended to information generated and verified by AI. It means that even if AI output is directly recorded on a blockchain, its correctness still requires separate verification.
* 'Black-box nature': Refers to a state where the decision-making process within an AI model cannot be observed or explained externally, making it impossible to trace how a particular conclusion was reached.
Chapter 3: Positioning of this 'Division Application'
The existing patent relates to the 'Appraisal Certification System (R)', a patent for evaluating verification information to guarantee the authenticity (trust) of all assets (including 'physical assets' like goods/products, 'non-physical assets' like data/content, and 'hybrid assets' like RWAs and digital twins that integrate both) using an evaluation model generated by machine learning, and outputting the verification results.
This division application...