In the artificial intelligence industry (AI) a new tool appeared to evaluate risks – Saif Risk Assessment, designed to increase the safety of AI systems. This interactive tool allows developers and companies to evaluate their security, determine potential threats, and strengthen the security system.
Last year, Google introduced the Secure AI Framework (SAIF), designed to help developers introduce AI safely and responsibly. The new tool available on the Saif.google website is aimed at the practical application of the SAIF principles, providing individual recommendations for the protection of AI systems.
Saif Risk Assessment offers users to answer questions related to training, configuration, and control of access to models, as well as protection against attacks. Based on the answers, the tool forms a report with identified risks, such as “data poisoning,” “injections of requests,” and other threats, and offers measures to eliminate them.
In addition, the latest Google report includes Saif Risk Map – an interactive map that demonstrates how risks can arise and be used at various stages of AI development. Users can study how certain threats affect safety and how they can be minimized.
The initiative is also supported by a coalition for safe AI (COSAI), uniting 35 partners. The coalition is working on creating solutions in the field of AI security, including risk management. Saif Risk Assessment is closely related to this task and aims to create a safer AI ecosystem.
For developers and companies who want to increase the protection of their AI systems, Saif Risk Assessment will become a useful tool. Additional information is available on the site saif.google, where updates are regularly published.