新聞正文

完整中文翻譯製作中，以下為原文。

Cycal Trust Inc. (Headquarters: Shibuya-ku, Tokyo; Representative Director: Tsuyoshi Sue; hereinafter "Cycal Trust") is pleased to announce that it has acquired a new domestic patent (No. 7872466) related to its "Appraisal Certification System (R)".This patent is the formal registration and numbering of a patent application (Application No. 2025-049833) that was announced as having received a notice of allowance (pending numbering) on April 21, 2026. The core of this patent lies in verifying the legitimacy of the "datasets (training data)" and verification information itself, which are input into and recorded by AI, *before* they are recorded, through a consensus mechanism by "AI Orchestration" (AI operated by multiple different entities). This verification process is then recorded on a distributed ledger (blockchain). In other words, it realizes the assurance and verification of "authenticity (trust)" through "AI Orchestration x Blockchain".Chapter 1: What is Happening Now?

~ The Question Beyond the "Four Corners of Othello" ~In June 2026, a policy was announced to establish leadership in the era of "Artificial Superintelligence (ASI)" by strategically controlling four domains: "AI Models," "Semiconductors," "AI Infrastructure," and "Robotics," referred to as the "Four Corners of Othello" (※1). These "Four Corners" are all domains that "strengthen" ASI. However, we believe there remains a critically important "question" here. That is, who guarantees that "the data forming the basis of ASI's judgments" is truly correct, after the "Four Corners" have been secured?

In an era where "AI Agents" make autonomous judgments and take actions, this "question" is rapidly becoming a reality. Recent industry surveys indicate that as "AI Agents" chain together "model calls, external data retrieval, and tool calls," their decision-making processes become increasingly opaque, leading to challenges described as a "black box within a black box" (※2).Chapter 2: "Another Layer" of the "AI Black Box Problem" ~The "AI black box problem" is often discussed in terms of the inability to explain the internal decision-making processes of models. This is the first layer that so-called "Explainable AI (XAI)" has addressed.However, in the age of "AI Agents," another layer is becoming more serious. This is the layer of "input and evidence authenticity (trust) and accountability," which asks: "Is the input data and verification information that AI receives as a premise for its judgments correct?

" and "Can we trace and verify which entity made the decision to accept or reject it, and on what grounds?

"This patent addresses the latter, the second layer. Instead of forcibly trying to peer inside the model, Cycal Trust proposes a practical solution to the "AI black box problem" by verifying, recording, and making traceable "the grounds on which AI makes its judgments."Chapter 3: Why is This a Problem Now?

~ "AI Agents" Operating with "False Verification Results" ~The lack of this second layer is already manifesting as concrete attacks. An industry survey on "AI Agent" security (2026) demonstrated a chain risk where verification agents return false "verified" information, causing downstream agents responsible for procurement and payment to transfer funds to an attacker's account (※3). This is not an "intrusion" into the system, but a cyberattack by a "highly autonomous AI (Claude Mythos-class AI)" that tricks the system into adopting "false information disguised as legitimate" as "genuine." This is a typical example of an "authenticity (trust) attack," which Cycal Trust has long pointed out.Furthermore, challenges remain with the reliability of cutting-edge models themselves. Public evaluations of hallucinations (false outputs by AI) (e.g., Vectara's "Hallucination Leaderboard") report that even new "inference" variants of major models such as "GPT," "Claude," "Gemini," and "Grok" show error rates exceeding 10% for new datasets (training data) (※4). "Being an intelligent model" does not necessarily equate to "being correct or honest."In government, the "Basic Plan for Artificial Intelligence (Draft)" released on June 19, 2026, highlights concerns about cyberattacks by "highly autonomous AI (Claude Mythos-class AI)" and the fundamental strengthening of "AISI (AI Safety Institute)" responsible for AI evaluation, traceability, and guardrails (※5). Cycal Trust interprets this trend as an increasing importance of technologies that guarantee "input and evidence authenticity (trust)."Chapter 4: Technical Configuration of This Patent ~ The Reality of "AI Orchestration" ~The reality of "AI Orchestration" patented here lies in its ability to cooperatively "direct" multiple AIs and verification models, rather than relying on a single AI. Specifically, it consists of the following elements:First, independent review by "Multi-AI." Multiple AIs operated by different entities independently review verification information, solving the "single point of failure problem" inherent in relying on a single AI.Second, pass/fail determination by weighting and thresholds. Each verification result is assigned a weight to calculate a "legitimacy evaluation score," and a "determination model" decides whether this score meets a preset threshold.Third, optimal allocation of verification. A "classification model" categorizes verification information, assigning the most suitable verification model for each category, and a "verification entity determination model" decides which entity will perform the verification.Fourth, recording the verification process. By associating verification information, verified information, and verification entities with identifiers and recording them on a distributed ledger (blockchain), it becomes possible to trace "which entity made the decision to accept or reject, and on what grounds?

" (ensuring accountability).This series of mechanisms is based on the concept of "preventing contaminated water from entering the pipes in the first place," rather than "purifying contaminated water after it has flowed."This can be said to be the solidification of the philosophy "Don't let AI learn lies," which Cycal Trust announced in April 2026, into registered rights.Chapter 5: Social ImpactIn an era where AI is integrated into the core of all decision-making, from finance and healthcare to autonomous driving, manufacturing, and administration, the ability to verify "whether the grounds for those decisions are correct" becomes the "fundamental basis of trust" for society as a whole.For example, in the semiconductor supply chain, counterfeit parts circulating with forged quality certificates pose a significant risk to economic security. "Data poisoning attacks," which inject fraudulent data into the "datasets (training data)" of "Generative AI" and "AI Agents," are also increasing internationally. The global damage from counterfeit goods is estimated to be approximately 645 trillion yen annually (※6), and much of this can infiltrate final products directly linked to human life, such as semiconductors, medical devices, automobiles, and aircraft. This patented technology provides a mechanism for "AI Orchestration" to examine authenticity (trust) before this data is recorded and learned.Furthermore, in Europe, obligations for high-risk AI systems under the "AI Act" will apply from August 2, 2026, requiring transparency, documentation, and human oversight (※7). The "DPP (Digital Product Passport)" also faces the challenge of ensuring the authenticity (trust) of the recorded data itself. While these are European regulatory frameworks and not legal obligations within Japan, Cycal Trust's patented technology can technically complement the authenticity (trust) and traceability (accountability) required by these regulations.Chapter 6: Implications for Three National Issues ~We believe that our accumulated patents, including this one, can function as a single "authenticity (trust)" layer that spans the "Four Corners" and addresses three national issues facing Japan.First, defense against cyberattacks that exploit "highly autonomous AI (Claude Mythos-class AI)." The essence of this threat lies in "authenticity (trust) attacks" that generate a large volume of "false evidence, records, and outputs" with legitimate formats, routes, and signatures, and have them adopted as "genuine" by humans or existing systems.We view the consensus verification by "AI Orchestration" as a verification layer that rejects such "legitimate-looking fakes" before adoption, complementing "preventative" defense (e.g., "Patching as a Service") (※8).Second, ensuring "trust by design" as advocated in the Liberal Democratic Party's "AI White Paper 2.0." We believe we can ensure and verify the authenticity (trust) of the recording and verification process end-to-end, preparing for "damage to trust" as called for in that document (※9).Third, realizing the "Next-Generation AI x On-Chain Finance Initiative." In a world where "AI Agents" are entrusted with automatic settlements, a layer that mechanically verifies the authenticity (trust) of transaction counterparties, assets, owners, and wallets is a prerequisite. This patent and its related patent group encompass the detection of fraudulent financial transactions, authentication of cryptocurrency transactions, and other identity verifications, and we see them as indispensable pieces supporting this prerequisite (※10).These three points are not separate, individual patent technologies. They all stem from a single "authenticity (trust)" layer that spans the "Four Corners."Chapter 7: Comment from Tsuyoshi Sue, Representative Director of Cycal TrustThe competition to "strengthen" the "Four Corners" is currently unfolding at the world's forefront. However, no matter how intelligent ASI becomes, society cannot trust its judgments if they are not based on "genuine data."What we are working on is not making AI "stronger" itself, but making AI "trustworthy"—building a single "authenticity (trust)" layer that spans the "Four Corners." We believe that this diligent step—verifying, recording, and making traceable "the grounds for its judgments?

" rather than "peering inside the model"—is the foundation for an era where AI bears the responsibility for societal decision-making.Cycal Trust will implement the "authenticity (trust)" foundation through "AI Orchestration x Blockchain" from Japan to the world.Chapter 8: About Cycal Trust(1) Company OverviewOur company develops the "Appraisal Certification System (R)," which uses verification through "AI Orchestration x Blockchain" to guarantee and verify the "authenticity (trust)" of all assets, "proof of identity," "proof of ownership of goods," "supply chain traceability certification," "environmental certifications such as carbon footprint and carbon credits," and other "CoP (Chain of Provenance)".The core patent group received the "Invention Award (Award for Meritorious Invention)" in 2025 in recognition of its achievements as a foundational patent. Our company is leading the international standard "ISO 26345" within "ISO/TC307 (WG8)".[Official Website]https://cycaltrust.co.jp/[Official YouTube]https://www.youtube.com/@cycaltrust_official(2) Affiliated Organizations・ "International Standards (ISO/TC307) WG8": Domestic Committee Member・ JIPDEC-hosted "Research Group for the Promotion of International Blockchain Standardization Activities": Member・ International Semiconductor Manufacturing Equipment Materials Association (SEMI): Related Member (Participating in Blockchain Working Group)・ Data Society Promotion Council (DSA): Supporting Member(3) Media Inquiries Regarding This Matter[Official Inquiry Form]https://cycaltrust.co.jp/jp/#contactChapter 9: Source of This Release(※1) "SoftBank Group Chairman Masayoshi Son Aims to Win the 'Four Corners of Othello' for ASI Group Strategy," Business+IT, June 24, 2026 (https://www.sbbit.jp/article/cont1/185888)(※2) Industry Survey and Analysis on the Opacity of AI Agents ("Black Box within a Black Box") (2026)(※3) Industry Survey on AI Agent Security (Memory Contamination, Indirect Prompt Injection, etc., 2026)(※4) Vectara "Hallucination Leaderboard" (Public Benchmark, 2026, as confirmed by our company)(※5) Cabinet Office "Basic Plan for Artificial Intelligence (Draft)" June 19, 2026 (AI Strategy Headquarters, Public Comment Stage)(※6) Frontier Economics. "The economic impacts of counterfeiting and piracy." International Chamber of Commerce (ICC), BASCAP, & INTA. 2017.(※7) European "AI Act (Regulation (EU) 2024/1689)" Obligations for High-Risk AI Systems: Applicable from August 2, 2026(※8) "To Defend Japan's Critical Infrastructure (Excerpt) Launch of 'Patching as a Service'," SoftBank Group Corp., June 16, 2026(※9) Liberal Democratic Party "AI White Paper 2.0" (Digital Society Promotion Headquarters, AI & Web3 Subcommittee, 2026)(※10) LDP Study Group on "Next-Generation AI x On-Chain Finance Initiative" (Newly established March 2026)