SHAREHOLDERPROPOSALS • PartnerandCollaboratewithExternalStakeholders.AWScollaborateswiththeacademiccommunityandotherstakeholders ontheresponsible use of AI/ML technologies. For example, through our participation in Partnership on AI, we have workedwithleading technology companies and organizations such as the ACLU, Future of Privacy Forum, and the MIT Initiative on the Digital Economy to advance public understanding of AI technologies and address opportunities and challenges with AI technologies to benefit people and society, focusing on areas such as ethics, fairness, inclusivity, and transparency. We are also active members of other multi-stakeholder organizations relating to AI, including The Organisation for Economic Co-operation and Development (OECD) working groups on AI. We also provide research grants through Amazon Research Awards and the joint Amazon and National Science Foundation Fairness in AI Grants program. • Require Customer Agreement to Acceptable Use Policy. As a condition to using Amazon Rekognition and every other AWS service, a customer(includinggovernmentorlawenforcementcustomers)mustaccepttheAWSAcceptableUsePolicy(the 111 “AUP”), which prohibits use of AWS’s services “for any illegal or fraudulent activity.” This includes the violation of any laws related to privacy, discrimination, and civil rights. AWS will suspend or terminate access to Amazon Rekognition if we determine a customer is violating our AUP or the AWS legal terms. • Enhanced Legal Terms. All customers using Amazon Rekognition must comply with the relevant AWS legal terms. In early 2020, prior to our implementation of the moratorium on police use, we spent significant resources and consulted with law enforcement customers, civil society groups, and other stakeholders to perform an extensive review of and update to our legal terms to require certain disclosures and practices around law enforcement use cases. For example, if a law enforcement agency uses Amazon Rekognition in connection with criminal investigations, AWS legal terms require it to publiclydiscloseitsuseoffacialrecognitionsystems,summarizethesafeguardsinplacetopreventviolationsofcivilliberties or equivalent human rights, and make the disclosure easily accessible; we also direct customers to resources made 112 available by the U.S. Federal Bureau of Investigation and Department of Justice in this area. In addition, if Amazon Rekognitionis used to assist in identifying a person, and actions will be taken based on the identification that could impact that person’s civil liberties or equivalent human rights, AWS legal terms require the decision to take action to be made byanappropriatelytrainedpersonbasedontheirindependentexaminationoftheidentificationevidence,andrequirethe agencytoensurethatsuchpersonnelreceiveappropriatetrainingontheresponsibleuseoffacialrecognitionsystems.113 Webelievethisframeworkstrikesabalancebetweenthebenefitsandrisksofuseoffacialrecognitionbylawenforcement andhelpsaddressconcerns around potential misuse. • Provide Customer Guidance on Best Practices and Acceptable Use. AWS provides guidance to customers on best practices forutilizingandanalyzingtheresultsfromusingfacialrecognitiontechnology.Forexample,inlinewiththeAWSlegalterms described above, AWS recommends that in public safety use cases human reviewers verify the system’s results and decisions not be made based on the system output without additional human review. AWS also recommends customers be transparent about the use of face detection and comparison systems in such use cases, including, wherever possible, informing end users and subjects about the use of these systems, obtaining consent for that use, and providing a 114 mechanismforendusersandsubjectstoprovidefeedbacktoimprovethesystem. AWSalsoprovidesguidanceto 115 customersontheresponsibledesign,deployment,anduseofMLsystems. Further, customerscanengageanAWSteam of experts in responsible ML to recommend and help apply existing use-case-specific best practices on the development, 116 deployment, and operationalization of responsible ML principles. As noted above, we have cross-functional experts fromengineering, science, product, legal, and policy backgrounds who establish processes and procedures to drive responsible use of AWS’s AI/ML services, including Amazon Rekognition. When we are approached by or become aware of customers with potential use cases that may implicate our AUP, these experts analyze the proposed use case and we have turned down customers whose proposed uses would violate our AUP. • Provide Reporting Mechanisms. AWS provides a website and e-mail address where any person can report suspected abuse, and AWS employs trained staff that review every report that is received. In the more than five years AWS has beenofferingAmazonRekognition,AWShasnotreceivedasinglereportofAmazonRekognitionbeingusedintheharmful mannerpositedintheproposal. 111 Available at https://aws.amazon.com/aup/. 112 Seehttps://aws.amazon.com/service-terms/ (Section 50.8.4). This term directs customers to example FBI statements, FBI privacy assessments, and the Facial Recognition Policy Development Template published by the U.S. Department of Justice’s Bureau of Justice Assistance; see also https://www.fbi.gov/ news/testimony/facial-recognition-technology-ensuring-transparency-in-government-use; https://www.fbi.gov/services/information-management/foipa/ privacy-impact-assessments/facial-analysis-comparison-and-evaluation-face-services-unit;https://bja.ojp.gov/sites/g/files/xyckuh186/files/Publications/Face- Recognition-Policy-Development-Template-508-compliant.pdf. 113 Seehttps://aws.amazon.com/service-terms/. 114 Available at https://docs.aws.amazon.com/rekognition/latest/dg/rekognition-dg.pdf. 115 Seehttps://d1.awsstatic.com/responsible-machine-learning/responsible-use-of-machine-learning-guide.pdf. 116 Seehttps://pages.awscloud.com/GLOBAL-aware-IND-AWS-ProServe-Responsible-ML-2021-reg.html. 2022ProxyStatement 85