Exam AWS Certified AI Practitioner AIF-C01 topic 1 question 30 discussion

Explicação: De acordo com a Generative AI Security Scoping Matrix, quanto maior o controle e customização de uma solução, maior é a responsabilidade da empresa pela segurança. Opção D representa o maior nível de propriedade: ao construir e treinar um modelo do zero, a empresa assume controle total sobre o modelo, os dados, o treinamento, a implantação e a segurança em cada etapa. As demais opções têm menos responsabilidade direta porque envolvem componentes terceirizados:

Building and training a generative AI model from scratch by using specific data that a customer owns: When a company builds and trains a model from scratch using its own or customer-specific data, the company has full control over the entire development process, including data collection, model training, infrastructure management, and deployment. As a result, the company assumes the most responsibility for security in all areas, from data protection to model security and the underlying infrastructure.

D: Building and training a generative AI model from scratch by using specific data that a customer owns. Explanation: When a company builds and trains a generative AI model from scratch, it assumes the most ownership of security responsibilities, including: Data security and compliance during training. Model development and training processes. Infrastructure and deployment security. Protecting the model from adversarial attacks. Ensuring ethical use of the model and safeguarding against bias and misuse. This approach provides complete control over the entire lifecycle of the AI solution but also places the greatest burden of responsibility on the company.

https://aws.amazon.com/ai/generative-ai/security/scoping-matrix/

D. Building and training a generative AI model from scratch by using specific data that a customer owns. In this scenario, the company has the most control over the entire development and deployment process. This includes: Data security: The company is responsible for securing the training data, which might contain sensitive information. Model security: The company needs to implement measures to protect the model itself, including securing the training process, model parameters, and deployment infrastructure. Operational security: The company is responsible for securing the deployment environment and monitoring the model for potential vulnerabilities. While the other options involve some level of security responsibility, they rely on third-party providers to a greater extent. This reduces the company's direct control over the security aspects of the solution.

D. Building and training a generative AI model from scratch by using specific data that a customer owns gives the company the most ownership of security responsibilities, as they are responsible for all aspects of the model's development and deployment.