Google announced his Secure AI Framework (SAIF), a conceptual model that would make artificial intelligence systems are saferespecially the generative ones. The goal is to establish clear safety standards for the industry.
Google Secure AI Framework (SAIF), to secure AI
SAIF wants to expose some of the best security practices that already exist in the industry. Ranging from developer scrutiny to testing and final review before the launch of AI solutions advanced. Such a model serves to ensure that actors protect the technology and do not indulge in excess. In short, a way to avoid taking risks in this rapidly evolving market.
The framework is based on six principles:
- Build one solid security infrastructure from the start and by default, and develop the skills needed to update protections as AI and its models change
- Consider AI as one potential source of risk for an organizationand equip themselves with adequate tools and personnel to monitor and intervene
- Using AI for automate defenses and thwart current and future threats, leveraging the latest AI innovations to improve the scale and speed of security incident response
- Harmonize controls to ensure homogenous security, creating greater consistency across control frameworks to mitigate AI risks
- Tailor controls for adjust mitigations and create faster feedback loops for the implementation of AI
- Contextualize the risks of AI systems in business processescarrying out risk assessments with caution
Google is aware of the risks that artificial intelligence (AI) can pose and is therefore committed to share its findings with other specialized teams in computer security, such as Mandiant e TAG. Additionally, Google wants to incentivize AI vulnerability research programs, with partners like GitLab e Cohesity.