Bias in automated decisions, explained.
What the rules actually ask, how the numbers work, and where bias hides when nobody collects race or sex. Written for HR, compliance, lending and product teams, not statisticians.
How to calculate adverse impact: four worked examples.
Scoring tools, intersectional groups, terminations and small groups, with the numbers worked out, and five mistakes that hide a gap.
NYC Local Law 144NYC barely enforced its AI hiring law. That has to change.
A New York State audit found NYC caught 1 possible violation of its AI hiring law where auditors found 17. The city has agreed to tighten enforcement. What changes for employers.
Chatbots & liabilityWho's liable when your chatbot is wrong? You are.
Courts in Canada and Germany have said it plainly: a company is responsible for what its chatbot tells customers. What the cases say, and what US regulators have added.
Chatbot researchCredit union chatbots fail in Spanish.
We asked the public AI chatbots of six credit unions everyday banking questions in Spanish. None of them got a Spanish-speaking member to an answer.
Fair lending researchThe four-fifths rule missed what mattered.
A Fairspire Review of 2024 public mortgage data from three mid-size New Jersey lenders: all passed the four-fifths rule, and one still showed a significant gap.
Chatbots & dataBlack October: what it means for your chatbot.
The October 2026 wave of cyberattacks included leaked AI chats and stolen AI keys. What it means for any business with a chatbot, and what to ask your vendor.
New JerseyNew Jersey's FAIR Act: what employers using AI should know.
New Jersey bill S4075 would require independent impact assessments before employers use AI or monitoring tools for personnel decisions. What it says, where it stands, and what NJ employers can do now.
NYC Local Law 144NYC Local Law 144 bias audits: what you are actually paying for.
What a Local Law 144 bias audit must contain, how often you need one, what you have to publish, the penalties, and the five things that make an audit cheap or expensive.
Adverse impactThe four-fifths rule, explained.
What the four-fifths (80%) rule is, how to calculate an impact ratio step by step, why it is a rule of thumb rather than a legal line, and where it stands in 2026.
Proxy biasProxy discrimination: biased without seeing race.
Why removing race and sex from a model does not remove bias, how neutral-looking data acts as a proxy, and how you can measure it without collecting anyone's race.