SHAREHOLDERPROPOSALS • EnhancedBoardOversight. Our Board has reviewed Amazon Rekognition, along with other programs, as part of numerous AWSbusinessreviews. In addition, our Nominating and Corporate Governance Committee has provided oversight on behalf of the Board over the human rights aspects of Amazon’s Rekognition technology, including specifically Amazon Rekognition’s facial recognition capabilities. These reviews focus on the actual operation and use of Amazon Rekognition, the potential concerns and abuses that critics have suggested could arise from the technology, and our actions to resolve or mitigate those risks and concerns. Under its charter, the Nominating and Corporate Governance Committee, which is comprised of directors with experience in emerging technologies and public policy, is given responsibility for overseeingandmonitoringtheCompany’spoliciesandinitiativesrelatingtocorporatesocialresponsibility,includinghuman rights and ethical business practices, and risks related to the Company’s operations and engagement with customers, suppliers, and communities. This Proposal Fails to Acknowledge or Address the Measures We Have Taken to Enhance Our Technology and Relies on Dated Claims and Mischaracterizations While we have been working to constantly enhance our AI/ML technology, including Amazon Rekognition, this proposal has relied on the same outdated assertions and mischaracterizations. For example, this proposal continues to mischaracterize AmazonRekognition as a surveillance program. In fact, Amazon Rekognition does not collect images for users to perform searches on and does not provide any photos or data for users to search or compare images against. Instead, the service can beusedtohelpidentify objects, people, text, scenes, and activities in images and videos, as well as to detect inappropriate content. Thus, the first element of this proposal, which requests a report on the extent to which Amazon Rekognition may target or surveil certain persons, is misleading, since Amazon Rekognition is not a surveillance technology; it does not target or surveil people any more than technologies like cell phones or cameras, which are also subject to potential misuse. Second,webelievethatthethird-partytestsfrom2018onceagaincitedbytheproponentdonotfairlyaddressAmazon’s Rekognition technology. While the advocacy group that conducted and published the tests has refused to published its data set, methodology, or results in detail, we have demonstrated that the group’s own description of its tests indicate that the technology was not used properly (for example, by using only an 80% confidence threshold that forces the service to return the most similar face even if there is not a clear match). When we have re-created their tests using the service correctly with a higher confidence threshold, the misidentification rate dropped to zero despite the fact that we used a much larger 117 dataset of faces. Theproposal requests that the Company prepare a report about the extent to which Amazon’s Rekognition technology mayendanger,threaten,orviolateprivacyandcivilrights,targetorsurveilparticulargroups,andcouldbemarketedandsold toauthoritarianorrepressivegovernments,aswellaspossiblefinancialrisks.Conversationsaroundresponsibledevelopment anduseofAI/MLsystemsarehappeningaroundtheworldamonggovernment,industry,academia,andothergroups. Amazonisanactiveparticipant and contributor to these conversations, and Amazon teams and subject matter experts are helping lead the industry on these very issues. As demonstrated above, we have conscientiously acted to review and address the concerns expressed in the proposal and transparently provided information regarding our actions to the public. In light ofourcommitmenttocustomertrust,privacy,andsecurity;thematerialbenefitstobothsocietyandorganizationsofAmazon Rekognition’s image and video analysis capabilities; and our ongoing transparency and efforts to address potential misuse of AmazonRekognition, the Board recommends that shareholders vote against this proposal. TheBoardofDirectorsrecommendsavote“AGAINST”thisproposalrequestingareportoncustomeruseofcertain technologies. 117 Seehttps://aws.amazon.com/blogs/aws/thoughts-on-machine-learning-accuracy/. In addition, in May 2020, a third party replicating the study of Congress membersreferencedintheproposalconfirmedthatwhenusedatthethresholdrecommendedforlawenforcement,AmazonRekognitionproducedno incorrect matches. See https://www.comparitech.com/blog/vpn-privacy/facial-recognition-study/. 86