A look at NIST's Facial Recognition Vendor Test program for facial recognition algorithms, which influences purchasing decisions of US agencies and business
Tom Simonite / Wired : Tweets: @mmullany , @packetswitchr , and @wired Tweets: Michael Mullany / @mmullany : Apart from Microsoft, Ever AI is the only non-Russian/Chinese company with a market leading score on NIST facial recognition benchmarks @packetswitchr : Apparently, @Microsoft has the best facial recognition algorithm, in one NIST's tests. https://www.wired.com/... @wired : A testing program run by the National Institute of Standards and Technology is vital to the facial recognition industry. Over 60 companies took part in the most recent rounds of testing, and the rankings are dominated by entrants from Russia and China. http://www.wired.com/...
Context & Ripple Effects
NIST's Facial Recognition Vendor Test has quietly become the procurement gatekeeper for face recognition: over 60 companies submit algorithms each round, and US agencies and businesses buy off the leaderboard. The rankings are dominated by entrants from Russia and China, with Microsoft holding the top score in at least one test and Ever AI the only other non-Russian/Chinese company with a market-leading result — which makes the benchmark as much a geopolitical scoreboard as a quality seal.
The Wired piece lands just before Slate reported that NIST builds its test sets from images of immigrants, visa applicants, abused children, and dead people collected without consent, and years before an [[a:838330|OIG report found Login.gov officials rejected facial recognition outright over algorithmic bias, breaking NIST rules]]. Together they frame the program's core tension: it steers federal buying power while its own data practices sit outside the consent norms it implicitly certifies.
First-order effects
- US agencies and businesses using the rankings as a buying filter effectively channel procurement toward Russian and Chinese vendors, unless they pay a premium for the two Western leaders, Microsoft and Ever AI.
- Microsoft and Ever AI get a durable sales asset no marketing budget can buy: a government-run benchmark score that functions as pre-approval for risk-averse public-sector buyers.
Second-order effects
- Every vendor selling face recognition into the US market is forced to treat NIST submission as a cost of doing business, concentrating R&D around beating the benchmark rather than buyer-specific needs.
- Slate's reporting on the test's unconsented image sources hands NIST a reputational problem that could push agencies to demand provenance and bias documentation alongside raw accuracy scores — a compliance burden that favors large vendors like Microsoft over smaller entrants.
Third-order effects
- A single national lab's test methodology is hardening into a de facto global standard for an industry whose leaderboard is majority Russian and Chinese, meaning US procurement policy is being set by whoever optimizes for NIST's datasets.
- If the pattern holds, accuracy leaderboards alone stop closing deals: the Login.gov rejection shows bias findings can override benchmark scores entirely, pushing the industry toward dual certification — NIST-style performance plus independent bias and data-provenance audits.
The trend: State-run benchmarks are becoming the invisible hand of AI procurement, with geopolitical concentration on the leaderboard forcing governments to weigh accuracy scores against bias and data-governance concerns.