Artificial intelligence detection programmes are being used by educators and human resource departments worldwide to catch cheaters.

The problem is that the same software has flagged Genesis in the Bible, the Declaration of Independence and Charles Dickens’s Oliver Twist as AI-generated — works that were penned long before AI was a gleam in anyone’s eye.

Bermudian barrister and former English teacher Zakkiyah Daniels thinks as many as 75 million students a month, worldwide, may be falsely flagged for turning in AI-generated writing.

She told The Royal Gazette: “An AI detector can generate a percentage in seconds, but the person behind that percentage may spend months proving they are human. In law, you have a right to know the case against you. A black-box score should never become a verdict.”

To tackle the problem, Ms Daniels is using AI to fight AI. She has created an app called BotOut that will scan your work for high-risk words.

It is not a humaniser that rewrites text. After a scan, users can click on each flagged word to see why it is risky, and get suggestions that keep their own voice.

Ms Daniels calls it the first “mis-AI-dentity risk checker”.

Part of the problem is that many AI large language models were trained on classics like Dickens or the Bible.

“The better you write, the more likely you are to get flagged as AI,” Ms Daniels said.

Conversely, Ms Daniels said AI engineers used casually written posts from internet sites like Reddit to represent normal human writing.

“It learns to write from us, but then it calls us the bots,” Ms Daniels said. “If your writing is unpredictable all over the place, has run-on sentences and bad grammar, then it will say, ‘that is human’.”

Turnitin, one of the most common AI detection tools used by colleges and universities, processes about 100 million student papers a year and admits a false-positive rate of up to 4 per cent, at the sentence level.

Non-native speakers, people with neurodivergence and Black people are at a greater risk of generating false positives.

Ms Daniels previously taught English at the Bermuda College, but was made redundant in 2022 after the Covid-19 pandemic.

She went to South East Asia on holiday and is still there.

The idea for the app came after she read an article about the number of students being falsely flagged. A number of top universities such as Stanford and Yale have stopped using the software on their students.

“I have always dabbled in tech,” she said. “For example, I started a smaller mental health app for Gen Zs, people born between 1997 and 2012.”

She decided to create something for honest writers.

Living in Vietnam, a country where the cost of living is relatively low, made it easier to focus on the project.

“I wake up, find a café with good coffee and start working,” she said. “Sometimes I work for 12 or 13 hours.”

The hardest part of the process was teaching herself the necessary AI tools. She started off vibe coding — using everyday language to prompt AI tools but upgraded to full-on AI software engineering.

There were some days where it took much patience not to throw her computer out of the window, she said.

BotOut is now on the market. There is a free version on its website. Users can also download a one time “quick fix” for $1.99. There is also a $9.99 monthly subscription option that includes 75 scans a month.

She is now developing a Chrome extension that shows AI-detection scores changing in real time.

Ms Daniels has also developed GenUWrite, a version for schools, employers and publishers. It records what triggered a concern, what evidence was reviewed and how a final decision was reached.

• To use BotOut, seebotout.app