AJL–Georgetown Safe Face Pledge (2018–2021)

On 11 December 2018, the Algorithmic Justice League and the Center on Privacy and Technology at Georgetown Law jointly announced the Safe Face Pledge — an industry-facing commitment framework designed to hold facial analysis technology vendors accountable for preventing weaponized, discriminatory, and lethal deployment of their products. The pledge launched with three inaugural corporate signatories — Robbie.AI, Yoti, and Simprints Technology — and was directed explicitly at a roster of major facial-recognition providers: IBM, Microsoft, Google, Facebook, Amazon, NEC, Megvii, and Axon.

The pledge framework

The Safe Face Pledge organized its demands around a four-part "SAFE" framework. Signatories committed to: Show value for human life, dignity, and rights — including an explicit prohibition on lethal applications of facial analysis technology and on selling to parties where lethal use was a foreseeable outcome; Address harmful bias through internal evaluation processes and by enabling independent third-party auditing of their systems; Facilitate transparency about how facial analysis tools were being deployed, particularly in government contexts where public scrutiny was otherwise absent; and Embed these commitments into legal contracts, procurement terms, and business practices rather than leaving them as aspirational ethics statements.

Joy Buolamwini framed the campaign's purpose as preventing facial analysis technology from "leading to collateral damage" — a direct reference to the documented consequences of inaccurate biometric identification in high-stakes enforcement contexts. The lethal-use prohibition would become the primary stated reason major vendors declined to sign: companies that supplied or anticipated supplying facial recognition to police, military, and border-control actors were not willing to publicly foreclose those markets.

Context: from Gender Shades to policy demand

The Safe Face Pledge was the advocacy capstone of a research-to-movement arc that had gathered force through 2018. AJL's Gender Shades study, co-authored by Buolamwini and Timnit Gebru, had documented error rates of up to 35 percent for dark-skinned women across commercial facial-analysis systems from IBM, Microsoft, and Megvii — systems already in deployment in law enforcement and border-control settings. The pledge translated that empirical finding into a minimum-standards demand: companies whose systems demonstrably failed darker-skinned and feminine-presenting faces were being asked to publicly commit to bias-remediation, transparency, and harm-avoidance practices, or to explain publicly why they would not. The demand structure — a public pledge rather than a regulatory submission — made declination visible in ways that not responding to policy consultations did not.

Adoption and refusals

Three technology companies signed the pledge at launch: Robbie.AI (a deep-learning firm providing real-time facial analysis services), Yoti (a British digital-identity company), and Simprints Technology (a biometric nonprofit focused on global-development applications in lower-income countries). The campaign attracted over 40 supporting organisations and more than 100 individual champions from AI ethics, civil liberties, and academic communities.

The major facial-recognition industry players — IBM, Microsoft, Google, Facebook, Amazon, NEC, Megvii, and Axon — did not sign. Microsoft and Facebook indicated they were reviewing the pledge; Google declined to comment. The explicit prohibition on lethal and law-enforcement uses was the publicly identified stumbling block: companies serving law enforcement and military clients were not willing to prohibit those applications.

Sunset and significance

AJL and Georgetown Law sunsetted the Safe Face Pledge in February 2021. Buolamwini's sunset announcement concluded that major vendors had "conclusively demonstrated that self-regulation is not enough to compel the comprehensive mitigation of abuses." Rather than framing the closure as failure, the announcement recast the pledge's run as a proof point: the industry had been given a clear pathway to voluntary accountability and had declined it at the level of the most harmful use cases, furnishing the civil society argument that binding regulation — not voluntary commitments — was the necessary instrument.

By February 2021 the regulatory environment had moved considerably. San Francisco (2019), Somerville, MA (2019), Oakland (2019), and Boston (2020) had enacted municipal bans on government use of facial recognition, and multiple federal legislative proposals — including the Commercial Facial Recognition Privacy Act and the Facial Recognition and Biometric Technology Moratorium Act — were circulating in the US Congress. Amazon, IBM, and Microsoft had announced voluntary suspensions of police sales in June 2020 following renewed scrutiny of algorithmic policing. The pledge had served less as a mechanism for voluntary compliance and more as a pressure instrument that made major vendors' declination visible and legible — clarifying for legislative actors whether corporate self-governance could be relied upon.

The Safe Face Pledge is the corpus's first major test case of the voluntary-corporate-commitment approach to AI accountability: an empirically grounded demand for explicit industry commitments whose sunset supplies the movement's primary documented rationale for why voluntary pledges were insufficient and binding legislative action necessary. It also marks the first campaign in which AJL directly deployed its audit research as leverage for a corporate accountability campaign — a pattern the organisation extended with the Freedom Flyers TSA opt-out campaign and the CRASH community-reporting project.