The Cognitive Reserve: How Labelbox's Alignerr Extracts Expert Minds for Algorithmic Warfare

Inside the Billion-Dollar Pipeline Converting Gig Workers Into MilitaryTraining Data.

Before the work, the harvest.

A glossy door in the cloud,
a promise printed in bright numerals—
up to, up to,
as if altitude were the same as arrival.

You bring your résumé like an offering,
your credentials like clean bones,
and a face the camera can hold steady
while an avatar named Zara
listens without listening.

Speak, expert.
Explain what you know.
Show how you weigh a claim,
how you name uncertainty,
how your eyes pause—
right there—
before you decide.

Somewhere, a funnel learns your shape.

Not hired? Still taken.
Hired? Still owned—
your judgment packed into boxes,
stacked in a warehouse of waiting,
Slack-lit, silent,
inventory that breathes.

A billion-dollar hand turns the crank:
sensors to shooters,
contracts to keywords,
biometrics to “operations,”
and the old spell called flexibility
over the newest kind of draft.

So let this be the threshold hymn:
Not fear, but clarity.
Not outrage, but pattern.
A vow to see what is being built
from the borrowed hours of the many—

and to ask, with steady voice,
when “human in the loop” is sold by the hour:
whose hands are on the wheel,
and who will carry the blame
when the loop closes.

The Pitch

Alignerr, powered by Labelbox, presents itself as a dream gig: flexible remote work training AI models, with rates "up to $150/hour" for domain experts. PhD holders, lawyers, engineers, scientists—all invited to join the "top 3% of AI trainers" shaping the future of artificial intelligence.

The application process is slick. Upload your resume. Verify your identity with facial recognition and ID scans. Then sit for a 15-minute video interview with "Zara," an AI interviewer that "delves into your background and expertise."

Sounds like an innovative hiring process. It's not.

It's a cognitive extraction operation hiding in plain sight.

The Players

Labelbox, Inc. is a San Francisco-based company valued at over $1 billion, backed by SoftBank Vision Fund, Andreessen Horowitz, and Databricks. They've raised approximately $189 million in venture funding.

In August 2022, Labelbox secured a $950 million contract with the U.S. Air Force for Joint All-Domain Command and Control (JADC2)—the Pentagon's initiative to connect "sensors to shooters" across all warfare domains at "the speed of relevance."

Alignerr is Labelbox's workforce platform—a gig economy marketplace for "AI trainers" who perform data labeling, model evaluation, and reinforcement learning from human feedback (RLHF).

The two are inseparable. Alignerr workers are the human cognition that Labelbox packages and sells to AI labs, Fortune 500 companies, and defense contractors.

What the Terms of Service Actually Say

Before you ever get "hired," you give a video interview to Zara. Here's what Labelbox's Terms of Service say about that interview:

"You hereby grant us an unlimited, irrevocable, worldwide, royalty-free, sublicenseable through multiple tiers, perpetual right to use, reuse, modify, display, publish and disseminate the Interview and your name, audio, photographic, video, and other likeness of you..."

"Alignerr and its successors, licensees and assigns shall own all right, title and interest in and to the Interview..."

"You understand and acknowledge that Alignerr, its successors, licensees, and assigns, have the right to edit the Interview Materials, to combine the Interview Material with other information or material, and to use the Interview, the Interview Materials, or any portion thereof to train or improve artificial intelligence technology..."

"You further understand and acknowledge that Alignerr has no obligation to use the Interview Material, nor any other obligation to you whatsoever, including any payment or compensation related to the Interview or the Interview Material, or any obligation whatsoever to admit you to the Alignerr Community, to hire you or to otherwise engage your services."

Read that again.

Every applicant—whether accepted or rejected—gives Labelbox perpetual, royalty-free, sublicensable rights to their video interview, voice, likeness, and demonstrated expertise. For free. Forever. With explicit permission to use it for AI training.

The "3% acceptance rate" they advertise? That means 97% of applicants donate free training data and receive nothing in return.

The Extraction Architecture

Here's how the system actually works:

Layer 1: The Application Funnel

Thousands of educated professionals—PhDs, lawyers, doctors, engineers, scientists—upload resumes and complete the Zara video interview. They explain their expertise, demonstrate their reasoning, reveal their domain knowledge.

Labelbox captures:

• How experts verbally explain complex concepts
• Speech patterns, facial expressions, micro-behaviors
• Domain-specific vocabulary and reasoning structures
• How humans signal confidence vs. uncertainty

Cost to Labelbox: Server hosting for Zara.

Value extracted: Massive corpus of domain experts demonstrating cognition, licensed in perpetuity.

Layer 2: Assessment and Profiling

Applicants who pass the initial screen complete additional assessments—language fluency tests, domain knowledge evaluations, skills assessments. More free cognitive labor generating training signal about what expertise looks like.

Layer 3: The Idle Reserve

Accepted workers join Slack channels and wait for projects. And wait. And wait.

Glassdoor reviews tell the story:

"Just another AI training website that overhires and doesn't deliver on work. Everyday they add between 15 and 50 new freelancers on Slack yet very few have any work to do."

"If they are stockpiling workers in hope of future work, it'd be nice to know. The stark lack of communication is disappointing."

"The idea behind the company was great but the work is highly unreliable... After months of waiting for the company to 'take off' with clients and late payment for work you manage to get, it's become apparent that the issue lies within the company itself."

The workers waiting in those Slack channels aren't failed hires. They're inventory. A vetted, profiled, credentialed cognitive reserve that can be activated when client projects arrive. Their availability is the asset Labelbox sells.

Layer 4: Active Extraction

When projects do materialize, workers get paid—but not the advertised "$150/hour." Glassdoor data shows reality:

• Average: $22-37/hour
• Typical range: $15-30/hour
• Compensation rating: 2.0 out of 5 stars (workers are not satisfied)

Workers generate training data. Labelbox sells it to clients at significant markup. Everything produced belongs to Labelbox under the independent contractor agreement.

Layer 5: Replacement

The AI systems trained on this human feedback will eventually automate the judgment these workers provided. They are training their own replacement—at gig rates, with no benefits, no job security, and no equity in the systems they help create.

The Defense Connection

This is where it gets darker.

Labelbox's $950 million Air Force contract is for JADC2—the Advanced Battle Management System designed to enable "real-time data sharing and machine-to-machine communication" across all warfare domains.

What does that require?

• AI systems that can make or validate targeting decisions
• Human-in-the-loop judgment for high-stakes military applications
• Domain experts who can evaluate machine decisions at scale
• A workforce that can be deployed rapidly without military employment overhead

Now look at what Alignerr provides:

• A vetted pool of domain experts across technical, scientific, and analytical fields
• Profiled reasoning patterns captured through video interviews
• Independent contractors with no employment protections or benefits
• Elastic capacity that can scale up or down based on contract flow
• All data owned by Labelbox, sublicensable to government clients

Alignerr isn't primarily a freelance marketplace. It's a cognitive reserve force for algorithmic warfare.

The workers sitting idle in Slack channels have been vetted, profiled, and catalogued. When a defense project requires human judgment at scale—validating AI targeting recommendations, evaluating threat assessments, providing RLHF for military decision systems—Labelbox can activate them.

Compare the models:

This is the privatization of military-cognitive labor through gig economy structures.

The Biometric Harvest

One more thing. Worker reviews consistently note the extensive personal data collection:

"They ask for too much personal data, bank details, facial scanners, ID pics, which I have not had in other much larger companies with high reputation and well structured offices."

The Privacy Policy confirms collection of:

• Biometric information (finger/face/voice prints)
• Sensory data (photos, videos, recordings of you and your environment)
• Geolocation data
• Professional and employment history
• Education records
• Characteristics of protected classifications (race, disability, medical condition, sex, gender, age, military status)

All of this is disclosed to "Service Providers" and "Affiliates" for purposes including "Recruiting, employing and/or engaging" workers and "Hosting and facilitating operations."

When your primary customer includes the U.S. Air Force, what does "facilitating operations" mean?

The Pattern

The structure excludes what it claims to include.

Alignerr claims to offer high-paying expert work. The structure delivers:

• Free training data from rejected applicants
• Below-market wages for accepted workers
• Comprehensive cognitive and biometric profiling
• A reserve labor force for defense applications
• All intellectual output owned by Labelbox in perpetuity

The workers think they're selling labor. They're being mined.

What You Should Know Before Applying

If you're considering Alignerr, understand what you're agreeing to:

1. Your interview is training data. Whether you're hired or not, Labelbox owns your video interview forever and can use it to train AI systems.

2. The advertised rates are marketing. Real pay is $15-35/hour for most workers, not "$150/hour."

3. Work is not guaranteed. You may wait months in Slack channels with nothing to do. Your availability is the product they're selling.

4. Everything you produce belongs to them. Under the contractor agreement, all work product is Labelbox property.

5. Your biometric data is collected. Facial recognition, voice prints, ID verification—all captured and retained.

6. Their primary customers include defense contractors. Your cognitive labor may ultimately serve military AI applications.

The Question

When a drone system needs human validation for a targeting decision, and that human is a gig worker getting $25/hour with no job security, no benefits, and a Terms of Service that grants perpetual ownership of their judgment patterns to a private company...

What does "human in the loop" actually mean?

What does accountability look like?

Who is responsible when the system fails?

Labelbox and Alignerr have built the workforce architecture of algorithmic warfare, hidden behind the language of "AI training" and "flexible work."

The workers are the product. The interview is the extraction. The idle reserve is the inventory.

And somewhere in San Francisco, a billion-dollar company is selling your cognition to the Pentagon.

Sources:

• Labelbox Terms of Use and Alignerr Applicant Privacy Notice (September 2025)
• Glassdoor reviews and salary data for Alignerr
• Labelbox funding data via Crunchbase, Tracxn, PitchBook
• Labelbox $950M USAF JADC2 contract (August 2022)
• Labelbox corporate materials and Alignerr Connect documentation
• U.S. Air Force Advanced Battle Management System (ABMS) program documentation