What is the anomaly trap, and how does AI misread people from certain ZIP codes?

Where I come from, by the numbers

I grew up in East Oakland, ZIP code 94603. According to the U.S. Census Bureau’s American Community Survey for 2020 to 2024, 31.6% of adults 25 and older there have no high school diploma, 14.7% hold a bachelor’s degree or higher, and about 3 in 1,000 hold a professional degree like a J.D.

Across California, fewer than 2 in 100 HR professionals hold a professional degree, based on the Census Bureau’s 2023 American Community Survey microdata. By the averages, someone from 94603 does not finish a professional degree or build a career in HR. I did both. I named the anomaly trap because I kept running into systems that were sure I could not exist.

How does the anomaly trap work?

An AI model answers from the patterns in what it was trained on. When it knows nothing about you, you get the default answer, written for the average person. When it infers who you are from indirect clues, such as a name, a school, or a ZIP code, it can price you by the averages of where you come from.

The same logic shows up in people. Economists at Boston University showed how employers can end up monitoring Black workers more closely than white workers, so the same mistake is caught and punished more often, even when the groups are equally productive (Cavounidis and Lang, NBER Working Paper 21612, 2015). California now regulates automated decision systems used in employment under FEHA, effective October 1, 2025, because the pattern is real.

Where does it show up in a job search?

• Resume screening. A filter trained on past hires rewards the schools, titles and phrasing of the people it already hired. A strong record in an unfamiliar format can score low.
• Salary tools. A pay estimate built from your current title and location can anchor you to where you have been instead of the role you are applying for.
• Advice. Ask a chatbot how to negotiate and you get advice written for the average candidate, which assumes the ask lands the same way for everyone. It does not, and the research on race and negotiation shows why.

People do it too. Being asked to sound, dress or present like the average hire is the everyday version, and code-switching has a measurable cost.

How do you get around it?

• Tell the tool who is asking. A prompt that states your role, level, market, and goal gets a better answer than one that leaves the model to guess.
• Check every legal claim against the source. AI states wrong things with confidence. A citation you can open is the test.
• Get a person in the loop. A referral puts a human reader between you and a screening filter. Here is how to get one when you know nobody at the company.
• Price yourself on your record, not your ZIP code. Market data for your role and level is the standard. Your background is an edge, not a discount. Start with the pay range.

What employers should take from this

If your organization uses software to screen, rank or score candidates, California’s rules treat that output like any other employment decision. Keep the records, test the results by group, and make sure a person with authority can overrule the tool. The full breakdown of how I test AI in HR work is on the AI and HR page.

Why I named it

Nobody was ever going to tell you how it works, so I did; now pass it on. The averages were wrong about me, and they are wrong about you. I use AI to compact the time. The thinking, and the read on you, stay with a person who knows where the system breaks.

McKinley holds a J.D. but is not a licensed attorney. Articles here are general information, not legal advice. For your specific situation, talk with an employment attorney.

Data current as of September 2026. Sources are linked where each figure appears.