Responsible AI has been a long-standing commitment at Amazon. From the beginning, we have prioritized responsible AI innovation by building safety, fairness, robustness, security, and privacy into our development process and educating our employees. We aim to establish and implement the necessary safeguards to protect our customers while making their lives better. Our practical approach to taking responsible AI from theory to practice, combined with our tools and expertise, enables AWS customers to effectively implement responsible AI practices within their organizations. To date, we have developed over 70 internal and external products, tools, and mechanisms to support responsible AI, published or funded over 500 research papers, surveys, and scientific blogs on responsible AI, and provided tens of thousands of hours of responsible AI training to Amazon employees. Amazon also continues to grow its portfolio of free responsible AI training courses for people of all ages, backgrounds, and experience levels.
Today, we are sharing progress on our Responsible AI efforts, including introducing new tools, partnerships, and testing that improve the safety, security, and transparency of our AI services and models.
We introduced new tools and capabilities to support adversarial testing (e.g., red team testing) and to safely build and extend generative AI.
In April 2024, we announced the general availability of Guardrails for Amazon Bedrock and model evaluation on Amazon Bedrock, making it easier to deploy safeguards, prevent harmful content, and evaluate models against key safety and accuracy criteria. Guardrails is the only solution offered by a major cloud provider that enables customers to build and customize safety and privacy protections for their generative AI applications in a single solution. It enables customers to block up to 85% of harmful content in addition to native protection from FMs on Amazon Bedrock.
In May we New AI service cards for Amazon Titan Text Premier Further supporting investments in responsible and transparent generative AI: AI Service Cards are a type of responsible AI documentation that provide customers with information in one place about intended use cases and limitations, responsible AI design choices, and best practices for deploying and optimizing the performance of AI services and models. We have created over 10 AI Service Cards to date to provide transparency to customers as part of a comprehensive development process that addresses fairness, explainability, truthfulness and robustness, governance, transparency, privacy and security, safety, and controllability.
AI systems may also have performance flaws and vulnerabilities that could increase the risk regarding security threats and harmful content. Amazon uses a variety of techniques, including manual red team exercises, to test AI systems and models, such as Amazon Titan. Red team exercises involve human testers probing for flaws in AI systems in an adversarial style, complementing other testing techniques, such as automated benchmarking against publicly available and proprietary datasets, and human-complemented evaluation against proprietary datasets. For example, we developed our own evaluation dataset of challenging prompts that we use to evaluate Titan Text's development progress. Because a single evaluation dataset is unlikely to give an absolute complete picture of performance, we test against multiple use cases, prompts, and datasets. Overall, Titan Text has undergone multiple iterations of red team exercises on issues such as safety, security, privacy, accuracy, and fairness.
We have introduced watermarks to help users determine whether visual content is AI-generated or not.
A common use case for generative AI is the creation of digital content such as images, videos, and audio, but to prevent the spread of misinformation, users need to be able to identify AI-generated content. Techniques like watermarking can be used to verify whether it comes from a particular AI model or provider. To reduce the spread of misinformation, all images generated by Amazon Titan Image Generator contain an invisible watermark by default. This is designed to be tamper-proof, and helps increase transparency around AI-generated content and combat misinformation. We have also introduced a new API (preview) in Amazon Bedrock that checks for the presence of this watermark and helps verify if an image was generated by Titan Image Generator.
Promoting collaboration between business and government on trust and security risks
Collaboration between companies, governments, researchers, and the AI community is essential to foster the development of safe, responsible, and trustworthy AI. Amazon is American Artificial Intelligence Safety Institute ConsortiumNational Institute of Standards and Technology (NIST)Amazon is working with NIST to establish a new measurement science to identify scalable, interoperable measurements and methodologies to advance the development of trustworthy AI. It is also providing $5 million in AWS computing credits to the institute for the development of tools and methodologies to assess the safety of underlying models. “Technology Agreement to Combat Deceptive Use of AI in the 2024 Elections” At the Munich Security Conference, this is a key part of our joint efforts to advance safeguards against fraud and protect the integrity of our elections.
As technology continues to evolve, we continue to explore new ways to facilitate information sharing and engagement between business and government. This includes: Thorn and all technology are human. We aim to design generative AI services to be safe and reduce the risk of them being misused for child exploitation. Frontier Model Forum Advancing science, standards, and best practices in the development of cutting-edge AI models.
We supported efforts to advance education by leveraging AI as a force for good to tackle society's greatest challenges.
At Amazon, we are committed to driving the safe and responsible development of AI for social good. We are seeing growing examples of how generative AI is helping address climate change and improve healthcare across industries. Brainbox AI, a pioneer in commercial building technology, launched the world's first generative AI-powered virtual building assistant on AWS to provide facility managers and building operators with insights to help optimize energy use and reduce carbon emissions. Gilead, a US biopharmaceutical company, is accelerating the development of life-saving medicines with AWS generative AI by understanding clinical trial feasibility and optimizing site selection through AI-driven protocol analysis leveraging both in-house and real-world datasets.
As we explore the transformative potential of these technologies, we believe education is foundational to realizing their benefits while mitigating their risks. That's why we provide education on the potential risks surrounding generative AI systems. Amazon employees: Tens of thousands of hours of training since July 2023It covers important topics like , risk assessment, and also dives deep into complex considerations surrounding fairness, privacy, and model explainability. As part of Amazon's “AI Ready” initiative to provide free AI skills training to 2 million people worldwide by 2025, we have launched new free training courses on safe and responsible use of AI in Amazon's Digital Learning Center. Courses include “Introduction to Responsible AI” for those new to the cloud on AWS Educate, and courses such as “Responsible AI Practices” and “Security, Compliance, and Governance of AI Solutions” on AWS Skill Builder.
Putting trust first to deliver groundbreaking innovation
As a pioneer in AI, Amazon is continually driving the development of safe, responsible, and trustworthy AI technologies. We are committed to establishing and implementing necessary safeguards while driving innovation on behalf of our customers. We are also committed to partnering with businesses, governments, academia, researchers, and others to enable breakthrough AI-generated innovations with trust as a top priority.
About the Author
Vashi Philomin He is the VP of Generative AI at AWS, where he leads the Generative AI effort, including Amazon Bedrock and Amazon Titan.
