Working Educators began as an organization fighting for racial justice in Philadelphia schools. That work led us to testing, to teacher leadership, and now to AI — because everywhere we look in education, the same patterns emerge. Technology that promises neutrality often encodes bias. Tools marketed as objective often harm the students already most marginalized.
AI detection is no exception.
The Research Is Clear
Multiple peer-reviewed studies have now documented what teachers have observed anecdotally: AI detection tools disproportionately flag writing by non-native English speakers and speakers of African American Vernacular English (AAVE).
Key Findings
- Stanford (2024): GPTZero and other detectors showed false positive rates 2-4x higher for ESL student writing compared to native English speakers.
- UC Berkeley (2024): Writing samples using AAVE features were flagged as AI-generated at significantly higher rates than samples using Standard American English with equivalent content.
- University of Maryland (2025): When the same content was expressed in formal vs. informal registers, the formal version was more likely to be flagged as AI — disadvantaging students who've been taught to write in academic English.
Why does this happen? AI detection tools look for patterns they associate with machine generation: predictable word choices, regular sentence structures, lack of idiosyncratic voice. But these same patterns can characterize the writing of:
- Students writing in their second (or third, or fourth) language
- Students who have been heavily coached on "academic writing" formulas
- Students deliberately avoiding the informal features of their home dialect
- Students using accessibility tools that standardize their output
In other words: the students most likely to be falsely accused of AI cheating are often students who've worked hardest to conform to academic writing expectations.
This Isn't New
Working Educators' racial justice work began with standardized testing. We documented how tests designed for middle-class white students systematically disadvantaged Black and brown children. We organized against using those tests to sort students into tracked programs.
The AI detection problem follows the same pattern:
- A technology is developed primarily by and for a narrow demographic
- It's marketed as objective and neutral
- It's deployed widely without adequate testing for disparate impact
- When bias is documented, the burden shifts to affected communities to prove harm
- Meanwhile, students face real consequences from flawed tools
We've seen this before. We're seeing it again.
What Teachers Can Do
Critical: Never use AI detection as the sole basis for academic discipline
This is the single most important protection against biased outcomes. A detection score is a prompt for conversation, not a verdict.
Before Using Detection Tools
- Know your student population. If you teach ESL students or students who speak AAVE, understand that detection tools are more likely to generate false positives for their work.
- Establish baseline writing samples. Have students write in class early in the term. This gives you genuine examples of their voice to compare against flagged work.
- Communicate with students. Be transparent about how and when you use detection tools. Students should know what they're being measured against.
When Detection Tools Flag a Student
- Start with curiosity, not accusation. Ask the student to walk you through their writing process. You'll often learn more from a five-minute conversation than from any detection score.
- Consider the student's context. Is this an ESL student? A student who received intensive writing instruction? A student using accessibility tools? These factors raise the likelihood of false positives.
- Document your reasoning. If you decide a flagged student did write their own work, keep notes. You may need to defend that judgment.
- If you decide academic integrity was violated: Base that conclusion on multiple forms of evidence, not detection alone. Process documentation, writing samples, oral explanation — these matter more than algorithm percentages.
Advocating for Change
- Push for bias testing. Ask your school or district whether detection tool vendors have published disparate impact data. Most haven't.
- Advocate for policy safeguards. Push for district policies that explicitly protect students from detection-only discipline.
- Talk to colleagues. Share what you know about bias in detection tools. Many teachers aren't aware.
A Note on Our History
Working Educators didn't start as an AI organization. We started as teachers in Philadelphia fighting for racial justice in our schools. We helped launch Black Lives Matter at School Week. We organized against discriminatory testing. We pushed for curriculum that told the truth about American history.
This work — documenting bias in AI detection — is a continuation of that mission. The technology changes. The pattern doesn't. Tools that harm marginalized students get deployed before they're ready, and teachers are left to manage the consequences.
We're done waiting for the vendors to fix it. We're documenting what we see, sharing what we know, and fighting for fair treatment of every student in our classrooms.
That's what Working Educators has always done.