
The Monthly Intelligence Report
The Rulebook and the Con: The July 2026 Intelligence Briefing
CARICOM Heads put artificial intelligence on their own agenda for the first time, COTED-ICT endorsed the UNESCO roadmap, and the Caribbean AI Task Force delivered its final report. Two things arrived in the same month with a fraction of the attention: a scam economy rebuilt around cloned voices and pointed at Caribbean families, and a marketing habit of presenting foreign models as local ones.
Note from the President
July gave this region more artificial intelligence policy than any month in its history. Our Heads of Government put AI on their own agenda in Gros Islet. COTED-ICT endorsed the UNESCO roadmap two days later. The Caribbean AI Task Force handed over its final report in St Augustine at the end of the month. Three documents, three rooms, one region. The Committee reads all of it below.
I want to flag two things that arrived in the same month and got a fraction of the attention.
The first is that the scam economy has been rebuilt around this technology, and it is aimed at us. Cloned voices of children calling their mothers. Fabricated endorsements from business people whose faces every reader of this Report would recognize. Fake relief appeals that will run the moment the next storm clears. If you take one practical thing from this issue, take the family verification word. Agree it this week, with your mother, with your father, with whoever would wire money if they heard you in trouble. It costs nothing and it works.
The second is a habit that has crept into how AI is sold in this region. Foreign models are being presented as local ones. Not always dishonestly, and I want to be plain that the Committee names no company and alleges nothing against anyone. But when a product is described as a Caribbean model, buyers and journalists have no way to check what that means, because we have never written down what it is supposed to mean. That is our failure, not theirs, and it is fixable. The Committee proposes a fix below.
And once again the Committee has insisted on carrying an item about an award The University of the West Indies gave me. I objected in June. I objected again in July. I have now lost the same argument twice in two months, which I am told is a Committee record.
Hurricane season enters its worst weeks. Check your preparations. Check on your parents.
Adrian Dunkley Founder and President, Caribbean AI Association
Feature
The Rulebook and the Con: The July 2026 Intelligence Briefing
By The CAIA Committee
July 2026 will be remembered as the month the Caribbean finally produced a shelf of AI policy. It should also be remembered as the month the region learned what it costs to have a shelf of policy and very little enforcement behind it.
We take the region first this month, because for once the region moved first.
Three rooms, one region.
The 51st Regular Meeting of the Conference of Heads of Government of CARICOM convened in Gros Islet, Saint Lucia from the fifth to the eighth of July. Artificial intelligence reached the leaders' own table, which it had never previously done, and it left with two outcomes. A coordinated regional approach to AI was named a priority alongside crime, food security, and climate finance. And Heads agreed to establish a regional blue-ribbon commission on artificial intelligence to guide policy that lets the region benefit from AI while managing its risks.
The Committee's assessment of that outcome is mixed, and we said so in our reporting at the time. Leaders' attention is the scarcest resource in Caribbean governance and AI now has it. That matters. But the Commission is the sixth regional body with an AI mandate, arriving into a room that already contains the CTU task force, the UNESCO roadmap process, the CARICOM Single ICT Space work, and several national strategies in draft. None of the five existing efforts has been told to stand down, and none has been named as the Commission's counterpart. Six mandates and one region is not coordination. It is the appearance of coordination, which is more dangerous, because it is harder to see.
On the seventh of July, CARICOM's COTED-ICT formally endorsed the UNESCO Caribbean Artificial Intelligence Policy Roadmap, built on consultation with more than a thousand stakeholders across the region. Endorsement is not legislation. Fifteen governments have now agreed on a direction. Not one of them has yet passed a statute that a citizen could rely on in court.
Then, on the twenty third and twenty fourth of July, The University of the West Indies and the Caribbean Telecommunications Union convened the first Caribbean Artificial Intelligence Forum at St Augustine. Ministers, diplomats, researchers, and industry sat in the same room for two days, and on the twenty eighth the Caribbean AI Task Force released its final report, Toward Harmonized AI Policies and Recommendations for the Caribbean. It sets five priorities: regional AI governance, data sovereignty and digital infrastructure, innovation and industry development, human capacity and AI literacy, and sustained multi-stakeholder engagement.
Read those five priorities against the PwC survey of a hundred and fifty four Caribbean board directors published this year, in which six per cent said their boards spend enough time understanding the impact of AI. The report asks the region to govern a technology that ninety four per cent of its directors are not yet giving time to. That gap is the actual work, and it will not be closed by another framework.
The global picture.
Three developments outside the region matter for Caribbean readers this month.
The second of August is now days away, and what applies on that date has changed. The European Parliament's Digital Omnibus on AI, approved in June, moved the compliance date for standalone high-risk systems under Annex III from the second of August 2026 to the second of December 2027. Members who spent the spring preparing for August have more time than they thought. What does apply from the second of August is Article 50, the transparency obligation, and it applies to any AI system placed on the European market. Systems that talk to people must say they are machines. Synthetic and manipulated content must be disclosed as artificially generated. Emotion recognition and biometric categorization systems must tell the people subject to them. A new prohibition also lands, covering applications built to generate non-consensual intimate imagery. Our Industry Council's final conformity clinic for members selling into Europe runs this month.
Second, the model market moved again. Anthropic released Opus 5 and overtook OpenAI on revenue, which is the first time the commercial ordering at the frontier has changed. OpenAI shipped an agentic work product. Google released three new Gemini variants aimed at production agents. For Caribbean buyers the practical consequence is not which company is ahead. It is that capability at a given price keeps falling, which means every cost estimate in a business case written before June is now wrong in your favour, and every procurement locked to a single supplier is now worse value than it looked when signed.
Third, Illinois became the first American state to require annual independent audits of the largest AI models, and Xi Jinping used the World AI Conference in Shanghai to set out a China-led alternative to United States governance efforts. The region will be asked, repeatedly and by both sides, to pick a lane. The Committee's position is that a fifteen-state bloc with no frontier capability of its own has more to gain from interoperability with everyone than from alignment with anyone.
The scam economy has been rebuilt, and it is pointed at Caribbean families.
The FBI's Internet Crime Complaint Center logged eight hundred and ninety three million dollars in AI-related fraud losses across 22,364 complaints in 2025. AI-enabled fraud grew twelve hundred per cent in a single year, roughly six times the growth rate of ordinary fraud. Elder fraud with an AI component accounted for about three hundred and fifty two million of that total. Investment fraud with an AI element accounted for about six hundred and thirty two million.
Those are American numbers because America counts. The Caribbean does not, and that is a separate problem. But the structure of the fraud tells us who is exposed here.
Three seconds of audio is enough to clone a voice. A voicemail greeting clears that bar. So does a church testimony posted to Facebook, a graduation speech, a WhatsApp voice note forwarded to a group. The scam that follows is a call from a child or a grandchild in trouble, needing money now, and it is devastating in this region specifically because our families are arranged across borders. A daughter in Toronto, a son in Florida, a mother in Portmore. Distance is normal, urgency is plausible, and the wire transfer infrastructure is already in the house.
The second pattern is the deepfaked local endorsement. Fabricated video of recognizable Caribbean business figures promoting investment schemes has been circulating since last year, and platform advertising review does not catch regional impersonation because it does not know these faces. The third is business email compromise, upgraded. The advice every regional bank gave its customers for a decade was to watch for poor grammar. That signal is gone. Attackers now write in the exact idiom of the target organization, then follow the email with a phone call in a cloned executive voice to confirm the instruction. The victim receives written and verbal confirmation from what appears to be the same person.
What actually works against all three is process, not vigilance. Agree a family verification word with the people who would send you money and the people you would send money to. In any organization moving funds, verify every payment instruction and every change of bank details by calling a number you already held, never a number supplied in the message. Require two authorizers above a threshold you set this week. And remove voice from your authentication entirely, because voiceprint verification in banking and telecommunications is now defeated by a technology available free to anyone.
The Caribbean AI Risk Management Council published its AI Risk Register this month, a register of more than five hundred AI risks with root causes and first controls, filterable by industry and jurisdiction. The fraud entries above are in it. It is free, and it is the most practical document any Caribbean board will read this quarter.
On calling a foreign model a local one.
The second caution is quieter and will do more long-term damage.
Across this region, products are being marketed as Caribbean AI, national AI, or sovereign AI when what has been built is an interface over a foreign general-purpose model. There is nothing wrong with building on foreign models. Almost everything useful in this region is built that way, including work the Committee admires. The problem is the claim, and the fact that no buyer, journalist, or minister in the Caribbean currently has any way to test it.
The Committee names no company and makes no allegation against anyone. We are describing a gap in the regional vocabulary, and gaps in vocabulary get filled by whoever is selling.
Here is the concrete harm. A ministry procures on the basis of data sovereignty and later discovers citizen data was routed through infrastructure in another jurisdiction. A minister cites a local model as evidence of national capability that does not exist, and budget that should have built capability goes somewhere else. And every firm doing the harder, slower work of genuine local building competes against claims it cannot match honestly.
The fix is a disclosure standard, and it is four lines long. Any product marketed as Caribbean, national, or sovereign AI should state the base model it is built on, the data it was trained or fine-tuned on, the jurisdiction where inference and data storage physically occur, and the specific local work performed, whether that is fine-tuning, retrieval over local sources, interface design, or integration. Four disclosures. A supplier doing real local work will answer them gladly, because the answers are the argument for buying.
The Association will publish a draft of this standard for member comment in September and will ask CARICOM procurement officers to adopt it as a tender condition. It is the smallest useful thing the region can do about AI misinformation, and unlike most of what was agreed in July, it requires no new body.
An honour at The UWI, in fuller measure.
We reported briefly in June that The University of the West Indies had conferred its Alumni Excellence Award on Adrian Dunkley. Our President objected then and objects now. The Committee returns to it because a single paragraph did not do justice to what the University was actually recognizing, and because the shape of the record is the argument.
The citation covers work across science, innovation, education, non-profits, and entrepreneurship, and the reason those five sit together is that one of them came first. Physics came first. Adrian is a physicist by training, a postgraduate of the Mona campus in climate studies and AI, and everything else grew out of the discipline rather than alongside it. The company was founded on the premise that Caribbean problems are measurement problems before they are software problems. The Genius Project, which put over two hundred Caribbean children aged five to eighteen through AI, machine learning and mathematics this year and awarded a million United States dollars in cash and prizes, teaches mathematics as the foundation because that is the order in which the subject actually works. This Association exists because he judged that policy without technical grounding produces documents rather than outcomes, and he has written a note in every one of its twenty four issues, including the two objecting to this item.
The University of the West Indies is the institution through which this region has told its young people what excellence looks like for three quarters of a century. Its honourees have been jurists, novelists, physicians, prime ministers. Placing a scientist who builds AI on that list tells every student on every campus that this is a life's work the region takes seriously, and that the route into it runs through the sciences rather than around them.
Congratulations, Mr President. Your objection is noted, filed, and overruled.
A closing thought.
The region wrote its rules this month and discovered, in the same weeks, that the people it needs to protect were already being robbed by the technology those rules describe. That is not an argument against writing rules. It is an argument about sequence.
A blue-ribbon commission will take months to seat. A harmonized policy framework will take years to legislate. A family verification word takes one conversation, and a four-line provenance disclosure takes one tender cycle. The region should do all four. It should not wait for the first two before doing the last two.
Emancipation Day falls on the first of August. The season is at its worst weeks. Prepare the house, agree the word, and read the register.
Originally published in The Monthly Intelligence Report, July 2026.
Read every issue of The Monthly Intelligence Report
One feature, one President's note, every month. Written by the CAIRA contributor bench from across the Caribbean and the diaspora.