AI Security Startup SherLOCK Provides Technical Support for the Formulation of 'AI Safety Evaluation Perspective Guide' in the AISI Healthcare Sector
SherLOCK Inc. participated in the AISI Healthcare Sub-Working Group and provided technical support for the formulation of the 'AI Safety Evaluation Perspective Guide', offering expertise in AI red teaming and data quality evaluation.
📋 Article Processing Timeline
- 📰 Published: April 3, 2026 at 20:40
- 🔍 Collected: April 3, 2026 at 12:30
- 🤖 AI Analyzed: April 21, 2026 at 03:39 (423h 8m after Collected)
SherLOCK Inc. (Headquarters: Minato-ku, Tokyo; Representative Director and CEO: Teresa Tsukiji; hereinafter "SherLOCK") provided technical support for the formulation of the "AI Safety Evaluation Perspective Guide" in the healthcare field as a member of the Healthcare Sub-Working Group (SWG) within the Business Demonstration Working Group (Business Demonstration WG) of the AI Safety Institute (hereinafter "AISI").
Leveraging its specialized expertise as an AI security startup, SherLOCK contributed to the systematization of specific evaluation methods for practitioners, particularly from the perspectives of security and data quality.
【AISI: Press Release】
https://www.ipa.go.jp/pressrelease/2026/press20260403.html
■ Background and Objectives
While generative AI is expected to bring significant innovation to the healthcare sector, it faces unique challenges such as hallucinations (generation of false information), advanced privacy protection, and ensuring security. As discussed at the "Hiroshima Global Forum for Trustworthy AI" in January 2026, the realization of "Trustworthy AI" is an international priority. SherLOCK has participated in this SWG and provided technical knowledge to create an environment where companies can guarantee safety from the development stage and balance business value with safety and peace of mind.
■ Main Support Provided by SherLOCK
In formulating this guide, SherLOCK contributed to the systematization of specific AI safety evaluation methods that practitioners can use immediately, focusing on the following areas:
Provision of expert knowledge on evaluation items and points from a security and data quality perspective using AI red teaming tests.
SherLOCK presented evaluation items and points for tests that assume the risk of intentional guidance to incorrect information via malicious prompts or the induction of information leaks. They provided technical expertise to build an evaluation process aimed at preventing serious incidents in medical settings.
They defined the risks of data quality used in RAG (Retrieval-Augmented Generation) and similar technologies compromising the reliability of the entire product, and organized and clarified the evaluation points to ensure the accuracy of medical and health-related information.
Advice on objective AI safety evaluation through third-party assessment.
Regarding AI safety and AI security, SherLOCK provided advice and case studies from a practical perspective on how AI red teaming tests conducted by third parties with specialized knowledge contribute to ensuring corporate reliability.
■ Features of this Guide
This guide has the following features so that even companies with few experts can easily utilize it.
5-phase evaluation following the AI lifecycle
Clearly defines the points to be evaluated in each phase: product design, model selection, implementation, verification, and introduction/operation.
10 multifaceted safety evaluation perspectives
Covers everything from universal AI safety evaluation perspectives to risks specific to healthcare. Explicitly states specific risks (e.g., direct harm to patient life and health) if evaluations are neglected.
Risk management support directly linked to practical work
Specific risk scenarios and countermeasures are organized, making them ready for immediate use by on-site personnel.
Leveraging its specialized expertise as an AI security startup, SherLOCK contributed to the systematization of specific evaluation methods for practitioners, particularly from the perspectives of security and data quality.
【AISI: Press Release】
https://www.ipa.go.jp/pressrelease/2026/press20260403.html
■ Background and Objectives
While generative AI is expected to bring significant innovation to the healthcare sector, it faces unique challenges such as hallucinations (generation of false information), advanced privacy protection, and ensuring security. As discussed at the "Hiroshima Global Forum for Trustworthy AI" in January 2026, the realization of "Trustworthy AI" is an international priority. SherLOCK has participated in this SWG and provided technical knowledge to create an environment where companies can guarantee safety from the development stage and balance business value with safety and peace of mind.
■ Main Support Provided by SherLOCK
In formulating this guide, SherLOCK contributed to the systematization of specific AI safety evaluation methods that practitioners can use immediately, focusing on the following areas:
Provision of expert knowledge on evaluation items and points from a security and data quality perspective using AI red teaming tests.
SherLOCK presented evaluation items and points for tests that assume the risk of intentional guidance to incorrect information via malicious prompts or the induction of information leaks. They provided technical expertise to build an evaluation process aimed at preventing serious incidents in medical settings.
They defined the risks of data quality used in RAG (Retrieval-Augmented Generation) and similar technologies compromising the reliability of the entire product, and organized and clarified the evaluation points to ensure the accuracy of medical and health-related information.
Advice on objective AI safety evaluation through third-party assessment.
Regarding AI safety and AI security, SherLOCK provided advice and case studies from a practical perspective on how AI red teaming tests conducted by third parties with specialized knowledge contribute to ensuring corporate reliability.
■ Features of this Guide
This guide has the following features so that even companies with few experts can easily utilize it.
5-phase evaluation following the AI lifecycle
Clearly defines the points to be evaluated in each phase: product design, model selection, implementation, verification, and introduction/operation.
10 multifaceted safety evaluation perspectives
Covers everything from universal AI safety evaluation perspectives to risks specific to healthcare. Explicitly states specific risks (e.g., direct harm to patient life and health) if evaluations are neglected.
Risk management support directly linked to practical work
Specific risk scenarios and countermeasures are organized, making them ready for immediate use by on-site personnel.