CXC Just Recorded Its First AI Misuse Cases: What 128 Flagged Students Mean for CSEC and CAPE
CXC's 2026 results are out, and for the first time the region's exam board is counting AI misuse as its own violation. The number is small next to 118,000 candidates. The direction it is moving in is not.
CXC's preliminary 2026 CSEC and CAPE results went live on August 18. Buried inside the pass-rate story was a smaller number that matters more for how the next exam cycle will run: 128 exam irregularity cases across the region, up from 80 the year before. For the first time, artificial intelligence misuse showed up as its own tracked category, sitting alongside unauthorised devices, collusion, and unauthorised materials as one of the top reasons a candidate's grade got cancelled.
Dr Nicole Manning, CXC's director of operations, named it directly when the results were announced: "the new one for us, artificial intelligence misuse." That single line marks a real shift. AI in Caribbean classrooms went from a policy question CXC was writing rules for, to a category of enforcement CXC is now actually applying.
The Numbers Behind the First AI Misuse Cases
CSEC irregularities climbed from 57 cases in 2025 to 106 in 2026, close to double. CAPE stayed roughly flat, 22 cases against 23 the year before. Combined across both exams, the region went from 80 flagged cases to 128. Any student confirmed to have committed one of these violations has their grade cancelled or is disqualified from the subject entirely, a decision that follows a candidate into university applications and scholarship files months later.
CXC has not published a subject-by-subject breakdown of exactly how many of those cases were AI-specific versus a phone in a bag or a formula sheet passed between desks. What is public is the ranking: unauthorised devices, collusion, and unauthorised materials remain the leading categories, with AI misuse named as the newest addition rather than a minor footnote. Put those two facts together and the pattern is plain. The 2026 sitting was the first full cycle to run under CXC's Standards and Guidelines for the Use of Artificial Intelligence in Assessment, the same cycle where AI misuse first became something the exam board had to name out loud.
🧠Why a clear rule still gets broken by accident
CXC's disclosure rule is not complicated: use AI, say so, on a form. But a sixteen-year-old finishing an SBA at midnight is not weighing that rule the way an adult would. Adolescent decision-making research consistently finds that teenagers process risk differently under time pressure and social comparison than they do in a calm classroom explanation of a policy weeks earlier. The rule was taught in September. The decision gets made in April, alone, against a deadline, with a friend's group chat open in another tab showing what everyone else already turned in.
That gap, between when a rule is taught and when it actually gets tested, is where most disclosure failures live. It is rarely a student trying to get away with something. It is a student who genuinely forgot the form existed by the time it mattered. Closing that gap does not need a scarier warning. It needs the habit built early enough, and repeated often enough, that filling out the disclosure becomes as automatic as saving the file.
Adolescent risk under pressureHabit formationDelay between rule and test
The Same Results Also Showed the Region Improving
It would be easy to read 128 irregularities and conclude the region's relationship with AI in exams is falling apart. The rest of the same results release argues otherwise. Regional CSEC Mathematics rose to a 42% pass rate, up four percentage points from 38% in 2025 and the third straight year of improvement after 36% in 2024. English A held steady at 80%. More than 118,000 students sat CSEC this year, and over 26,000 sat CAPE. Manning's own comment on the maths gain was blunt: "Any improvement in math is a good improvement, so it's important for us to be happy about that. Improvement is improvement."
Both things are true in the same dataset. Pass rates are climbing, and enforcement against misuse is climbing with them. That is not a contradiction. It is what happens when a tool spreads through a student population faster than the rules around it get taught, tested, and finally enforced.
By the numbers
Caribbean Examinations Council, 2026 May-June exam results, reported August 18, 2026
Anguilla Already Solved the Part the Rest of the Region Is Still Guessing At
While the wider Caribbean was catching its first AI misuse cases, Anguilla was already a year into running the answer. In 2025, Anguilla became the first CARICOM member state to adopt a National Generative AI in Education Policy, running through 2030, after signing a memorandum of understanding with CXC to bring AI into classrooms in a structured, disclosed way rather than leaving individual schools to work it out alone. CXC's Responsible Generative AI Policy Framework, built on the back of that partnership, has since gone out to secondary schools across Antigua and Barbuda, Barbados, Belize, Dominica, Grenada, Guyana, Jamaica, Saint Lucia, Saint Vincent and the Grenadines, and Trinidad and Tobago.
The lesson from Anguilla is not that a national policy document eliminates misuse. It is that a territory which taught the disclosure rule early, consistently, and with actual classroom time behind it walks into an exam cycle with fewer surprises than one that handed teachers a framework and hoped the habit would form on its own. Every territory now holding CXC's framework is in a race against the same April deadline pressure described above. Anguilla just had a head start.
"Mi tell mi students di same ting every term: nuh wait til di night before fi find out weh di rule seh. By di time yuh a rush, yuh nuh have time fi think, yuh just click submit. Di ones weh nevah get flag, dem build di habit early, dem know exactly weh dem use AI fah before dem even start di assignment." - Andre Facey, AI Literacy Mentor, The Genius Project
What This Means If You're Sitting CSEC or CAPE Next
Closing the gap between using AI and disclosing it correctly does not require a new subject on the timetable. It requires a small set of habits, built well before the deadline that makes them hard to remember.
- Know the disclosure rule before you open the assignment, not before you submit it. Under CXC's Standards and Guidelines, any AI use on an SBA or IA needs a Disclosure Form and an originality report. Read what that actually requires in September, not April.
- Log AI use as you go. A running note of exactly where and how AI helped is far easier to write honestly at the time than to reconstruct from memory the night before a deadline.
- Use AI to check your reasoning, not to produce your final answer. Feeding it a past-paper question and asking it to quiz you back builds understanding you can defend under exam conditions. Asking it to write the answer builds nothing you can explain if a teacher asks you to.
- Ask when the rule is unclear. A five-minute question to a teacher before submission beats a grade cancellation after it.
- Build the habit somewhere it costs nothing to get wrong first. A free programme, a mentor, or a low-stakes project is a better place to make your first AI-disclosure mistake than a CSEC SBA worth a real grade.
How The Genius Project Fits In
That last point is the one The Genius Project exists to solve. The programme is tuition-free for Caribbean youth and teaches AI literacy, including disclosure habits, months before a student's next SBA deadline puts pressure on the decision. It was founded by Adrian Dunkley, who also built StarApple AI, the first AI company launched in the Caribbean, and who is widely regarded across the region as its leading voice on how AI should actually be taught rather than just permitted.
The Genius Project sits inside a wider Caribbean AI network built around that same goal. The Caribbean AI Association works on the regional policy side of exactly this problem, pushing for shared standards across territories rather than each one improvising alone. StarApple AI supplies the applied expertise that shapes what gets taught in the programme itself. And for more of Adrian Dunkley's work across the region's AI policy conversation, see adriandunkley.net.
What Comes Next
The gap this article opened with, 128 flagged cases against 118,000 CSEC candidates alone, is still small in relative terms. What matters is the direction. This was the first cycle where AI misuse existed as a category CXC could point to at all. Next year's results will show whether that number keeps climbing as more students use AI without being taught the disclosure habit, or starts falling as CXC's Responsible Generative AI Policy Framework reaches classrooms that have not seen it yet.
Anguilla's head start says the second outcome is achievable, not theoretical. It just takes the same thing it always takes: teaching the habit early enough that it survives contact with an actual deadline.
See also: CXC's AI Exam Rules Just Had Their First Real Test, CXC Rewrote the SBA to Stop AI Cheating, and Your CSEC Results Are Coming: Use This Six-Week AI Window.