The Securities and Exchange Commission has told Amazon that two shareholder proposals seeking limits on the company’s face recognition software are appropriate, and should be put up for a vote at Amazon’s annual meeting.
The assessment adds regulatory weight to a growing chorus of researchers, ethicists, and Amazon employees who have raised concerns about the software and asked the company to halt sales to government customers, citing privacy harms and racial and gender bias.
The tech giant’s lawyers had argued that the shareholder proposals should be excluded from a vote at the company’s meeting in May because they are “an insignificant public policy issue for the Company” and do not relate to its ordinary business.
The SEC’s Division of Corporation Finance, however, disagreed. On Friday, lawyers for Amazon filed a last-ditch request asking the SEC to reconsider its judgement, but on Tuesday the Division denied that request.
“As a result, there will be two shareholder proposals regarding facial recognition on the ballot at Amazon this year,” said Michael Connor of Open Mic, a nonprofit focused on corporate accountability that helped organize the effort.
One of the shareholder proposals asks Amazon to stop selling the software, Amazon Rekognition, to government agencies unless it can use independent evidence to show that it doesn’t endanger human rights. The other proposal asks the company to consult with outside experts to assess the software’s risks to privacy and civil rights, and how to mitigate those risks.
A separate letter to Amazon published on Wednesday and signed by at least 50 leading artificial-intelligence researchers echoed the shareholders’ demands, and reviewed findings about the software’s problems with racial and gender bias. The signatories included researchers from Facebook, Microsoft, and Google, as well as a former principal scientist at Amazon Web Services, Anima Anandkumar.
Amazon Web Services markets Rekognition to a wide range of customers; it helped ID celebrities during a royal wedding telecast last year, for instance. The software has also been used by at least two U.S. police departments, and the FBI is using it to search for faces in mountains of video surveillance as part of a pilot that began in early 2018. In October, The Daily Beast reported that Amazon has pitched Rekognition to U.S. Immigration and Customs Enforcement.
“The controversy surrounding the technology threatens the relationship of trust between the Company and its consumers, employees, and the public at large,” the shareholders told the SEC in a February analysis, one of a volley of letters sent by their lawyers and Amazon’s. The Sisters of St. Joseph of Brentwood filed the resolution as shareholders and members of the Tri-State Coalition for Responsible Investment, which represents a group of investors with over $1.32 billion worth of total assets, according to a statement.
“The company’s products include a number of other products—Alexa, Ring, and Eero—that will face a spillover effect if Amazon’s status as a trusted company is breached by concerns about privacy and surveillance,” the shareholders wrote. “Moreover, in addition to the Company’s unique exposure to risk by virtue of it being a business operating in the technology sector, it also has a product pipeline and pending patent applications which demonstrate the trajectory of the company is on a collision course with just such concerns.”
An Amazon representative declined to comment on the proposals or on the researchers’ open letter, but pointed to company blog posts that tout the benefits of face recognition software, including fighting child sexual abuse, finding missing children, and improving content moderation. Amazon contends that its policies allow it to suspend customers from using its services if they are found to be using them in unlawful ways, and that it has received no reports of misuse of the software by police.
Demands to stop government sales
In the absence of any federal law governing the use of facial recognition software, Amazon has faced a volley of questions from privacy advocates and lawmakers over Rekognition’s use by law enforcement.
Last year, employees at Amazon and Microsoft in Seattle met with groups representing Muslims, African-Americans, immigrants, and other minority communities to hear firsthand stories of the impacts of targeted surveillance. The groups urged the companies to not sell face recognition technology to government agencies, according to Shankar Narayan, the Technology and Liberty Project director of the ACLU of Washington, which helped to facilitate the meetings.
In November, 450 Amazon employees also signed a letter to CEO Jeff Bezos calling for the company to stop selling Rekognition to police departments and to institute employee oversight for ethical decisions.
In January, a letter signed by the American Civil Liberties Union (ACLU) and dozens of human rights groups asked the company to cease government sales of the software, warning of its harms on over-policed and over-surveilled communities in particular.
“In a world with face surveillance, people will have to fear being watched and targeted by the government for attending a protest, congregating outside a place of worship, or simply living their lives,” the letter said.
The privacy risks that come with face recognition software are dystopian, but adding in the possibility of bias and error that can come with AI-driven systems only enhances those harms. In a peer-reviewed study published in January, MIT researchers showed that Amazon Rekognition had more difficulty identifying the gender of female and darker-skinned faces than similar services by IBM and Microsoft. The ACLU also raised concerns of racial bias in Rekognition after conducting tests last year.
Amazon has challenged what it called “misinterpreted results” of the research on Rekognition, arguing that the testers used an improper “default confidence threshold.” The open letter on Wednesday picked apart Amazon’s objections, calling its response to the peer-reviewed research “disappointing.”
The tech giant is one of a growing crowd of firms selling face recognition to law enforcement around the world—as well as to stores, casinos, airports, schools, and sports arenas. Still, Amazon’s stance on face recognition stands in contrast to that of the other large cloud companies.
Last year, Google announced it would cancel a Pentagon contract following employee protests, and has said it will not sell a face recognition product until risks could be mitigated. Microsoft vice president Brad Smith has been vocal about the need for regulation.
In a November interview, Bezos defended the company’s government sales, and argued that some misuse of certain technologies may be inevitable, but that should not be a reason to curtail their development.
“I feel that society develops an immune response eventually to the bad uses of new technology, but it takes time,” Bezos told Wired’s Steven Levy.
Still, he added, “I worry that some of these technologies will be very useful for autocratic regimes to enforce their goal.”
Calls for laws, including by Amazon
After the shareholder proposals were filed in January, Amazon lawyers mounted a significant legal effort to block them, asking the SEC for a “no action” letter that would allow the company to exclude the petitions from the proxy statements it gives to investors.
In its response to Amazon on March 28, the SEC’s Division of Corporation Finance said it was “unable to conclude that the proposals are not otherwise significantly related to the Company’s business.”
“In our view,” it added, “the Proposals transcend ordinary business matters.”
The SEC cannot mandate the inclusion of the proposals in the shareholder meeting. But the agency’s response gives significant legal backing to the shareholders, should Amazon choose not to include the proposals at the annual meeting.
Amazon, like many companies, has sought to block at least eight shareholder resolutions this year. The tech giant did not challenge a climate change-related proposal submitted by employee-shareholders in December that calls for a public report on how the company is cutting its use of fossil fuels.
Shareholder votes are non-binding, but they can help draw attention to problems, elicit on-the-record responses from the board of directors, and put pressure on the company to change.
Some companies like Axon have established ethics boards to self-police new products like face recognition, but privacy advocates insist that only stronger regulation and more transparency can limit the risks to the public. As the Amazon investors note, Microsoft’s Smith has also joined the calls for new face recognition laws.
In a February blog post, Amazon Web Service’s Global Public Policy VP, Michael Punke, also outlined specific areas where government regulation was needed, especially around police uses of face recognition.
One proposed law in Washington State that’s backed by Microsoft would mandate that signs be posted in public places where facial-recognition is in use, and require the government to obtain a court order when looking for specific individuals. But some privacy advocates have called for a stricter set of rules—or a broader moratorium on the technology.
In his blog post, Amazon’s Punke pushed back on that idea. “New technology should not be banned or condemned because of its potential misuse,” he wrote.
“Instead, there should be open, honest, and earnest dialogue among all parties involved to ensure that the technology is applied appropriately and is continuously enhanced.”