How AWS can help mission-focused organizations comply with the White House National Security Memorandum on AI

On October 24, 2024, the White House released a National Security Memorandum (NSM) on Artificial Intelligence (AI), which focuses on ensuring US leadership in developing advanced AI technologies. Amazon Web Services (AWS) is uniquely positioned to address the critical needs of the defense and national security customers in advancing their AI capabilities. Our comprehensive suite of AI and high performance computing (HPC) capabilities offers flexible and robust solutions to meet the NSM’s goals and empower national security missions.

For example, Amazon Bedrock, AWS’s foundational offering for generative AI, provides a versatile service for building multipurpose capabilities. Its modular, microservices-oriented approach allows teams to quickly assemble tailored solutions by mixing and matching specific capabilities. This architecture enables the development of highly scalable, flexible, and adaptable generative AI applications, with each component deployable and scalable independently. The versatility of Amazon Bedrock’s AI foundation, combined with the agility of a microservices approach, empowers organizations to rapidly respond to changing needs and unlock new opportunities through innovative generative AI solutions. Let’s take a closer look at how the company’s initiatives align with the memorandum’s main focus areas.

Leading the world in trustworthy AI

AWS actively supports the US government’s goal to lead in safe and responsible AI innovation, as outlined in the NSM. In February 2024, AWS announced a collaboration with the U.S. Artificial Intelligence Safety Institute Consortium (AISIC) at the National Institute of Standards and Technology (NIST), to establish new measurement science for AI. This partnership aims to develop proven, scalable, and interoperable measurements and methodologies that promote responsible AI development and use.

Through this collaboration, AWS is helping to advance rigorous evaluation of AI systems for potential safety, security, and trustworthiness risks before real-world deployment. Our $5 million commitment in compute credits to AISIC enables the development of tools and methodologies for organizations to evaluate their foundation models’ safety. AWS is particularly focused on developing evaluation methodologies for large parameter models and improving pre-deployment testing by addressing domain-specific risks.

AWS’s commitment to security and compliance directly aligns with the NSM’s trustworthiness goals. Our AI services are built on a foundation of stringent security measures, ensuring that defense and national security customers can develop and deploy AI models with confidence. Amazon SageMaker provides a secure environment for building, training, and deploying machine learning (ML) models, while Amazon Bedrock offers a selection of foundation models that have undergone rigorous testing for safety and reliability testing.

Strategic partnerships for enhanced trust

AWS’s strategic partnerships with leading AI organizations like Anthropic and NVIDIA contribute additional layers of trust. Our strategic collaboration with Anthropic, a company founded on developing safe and reliable AI systems, exemplifies this commitment. Anthropic’s Constitutional AI, which aligns language models with high-level normative principles, provides additional guardrails around generative AI applications. We also work with partner network companies such as Palantir to leverage these models for national security use.

Further, our 13-year partnership with NVIDIA, beginning with the launch of the first GPU cloud instance on AWS, further strengthens our capabilities. The collaboration on Project Ceiba, one of the world’s fastest AI supercomputers hosted exclusively on AWS, leverages NVIDIA’s cutting-edge GPU technology. This partnership enables advancements in healthcare, drug discovery, and large language models (LLMs) by combining NVIDIA’s hardware with AWS’s scalable cloud infrastructure.

Harnessing AI for national security missions

AWS actively supports the US government’s directive to harness powerful AI capabilities with appropriate safeguards for critical national security objectives. In addition to our years to work supporting the US government, we also partnered with Anthropic and Palantir to provide US intelligence and defense agencies access to Anthropic’s powerful AI models like Claude 3 and 3.5 within Palantir’s secure AI Platform hosted on AWS. This integrated solution harnesses AI to rapidly process vast data, identify patterns, streamline document review, and provide data-driven insights to inform mission-critical decisions – all while meeting stringent security protocols like DISA Impact Level 6 accreditation. By operationalizing advanced AI securely in classified environments, AWS enables government agencies to gain an asymmetric advantage in areas like intelligence analysis, decision-making, and operational efficiency for their most sensitive security missions.

These efforts enable effective and responsible AI adoption in the public sector, aligning with the national security objectives outlined in the memorandum. The NSM emphasizes the need for rapid AI adoption in national security. AWS’s scalable infrastructure and ready-to-use AI services enable defense and national security customers to quickly implement AI solutions across various mission areas. Our tools support the analysis of vast datasets, enhance cybersecurity efforts, and improve decision-making processes – all crucial for national security operations.

Promoting global AI governance

AWS demonstrates a strong commitment to promoting AI governance through various initiatives and partnerships. Our active participation in international standards through organizations like International Standards Organization (ISO) and NIST, contributes to creating global AI governance frameworks. By implementing robust security measures across our AI services – including Amazon SageMaker, Amazon Bedrock, and Amazon Q, we maintain compliance with regulatory requirements and industry standards, directly supporting the NSM’s goal of cultivating a stable and responsible framework for international AI governance.

AWS’s responsible AI principles and initiatives closely align with this objective through our comprehensive suite of tools that address multiple dimensions of responsible AI. This framework enables defense and national security customers to expedite AI adoption responsibly through eight key pillars:

1. Safety: Amazon Bedrock and Amazon Q Business feature multi-layered safeguards including guardrails, content filters, and contextual grounding checks to promote responsible AI use.
2. Fairness: Amazon SageMaker Clarify provides sophisticated tools for detecting and mitigating bias in ML models, offering comprehensive evaluation capabilities across various domains and model types.
3. Explainability: SageMaker Clarify delivers explainability tools for traditional deep learning models while exploring innovative approaches for LLMs like chain of thought reasoning and trace functionality.
4. Privacy and Security: AWS implements a shared responsibility model, offering robust encryption, access control, and security features across its services.
5. Controllability: AWS provides comprehensive control tools including Amazon Identity and Access Management (IAM), Amazon CloudWatch, AWS CloudTrail, and Amazon Augmented AI, enabling secure access control, monitoring, auditing, and human oversight of AI systems.
6. Veracity and Robustness: AWS implements Retrieval-Augmented Generation (RAG), performance monitoring, and human-in-the-loop capabilities to enhance reliability and reduce hallucinations in AI applications.
7. Governance: SageMaker Role Manager, SageMaker Model Cards, SageMaker Model Dashboard, and AWS Audit Manager support responsible AI implementation through role-based access control, documentation, and compliance monitoring.
8. Transparency: AWS promotes transparency and mitigates misuse of AI-generated content through service cards, comprehensive auditing capabilities, and features like invisible watermarking.

These tools work in concert to address key dimensions of responsible AI, enabling agencies to develop and deploy AI systems that are ethical, trustworthy, and aligned with societal values.

Why AWS is uniquely able to support the NSM

In summary, AWS’s active engagement in initiatives supporting key priorities outlined in the NSM – including AI safety testing, responsible AI development for the public sector, and promoting international norms for trustworthy AI – demonstrates its strong alignment with the US government’s strategic goals. AWS’s comprehensive tools and Amazon Bedrock’s flexible architecture provides a solid foundation for agencies to develop and deploy AI systems that are ethical, trustworthy, and aligned with societal values, making AWS uniquely positioned to support the implementation of the NSM.

Contact your AWS public sector team or your account team to understand how AWS can help you with your AI adoption.