CARICOM Put AI on the Agenda for the First Time at Its 51st Summit. The UN's Scientists Say Nobody Can Guarantee It Is Safe.

Two summits happened in the same week of July 2026, on opposite ends of the process that decides how artificial intelligence gets governed. In Geneva, the United Nations convened governments, technology companies, and civil society for its Global Dialogue on AI Governance, days after a panel of forty scientists warned that nobody can currently guarantee the safety of the most advanced AI systems. In Gros Islet, Saint Lucia, fifteen CARICOM Heads of Government met for the 51st time and, for the first time in the Conference's history, placed artificial intelligence on their own agenda rather than routing it through telecommunications ministers. Both events produced statements. Neither produced a rule a Caribbean compliance officer can point to on Monday morning.

Two Summits, One Week

The timing was not coordinated, which is itself worth noticing. On 1 July, the UN's Independent International Scientific Panel on AI released its preliminary report. On 6 and 7 July, the Global Dialogue on AI Governance opened in Geneva, co-chaired by Ambassador Rein Tammsaar of Estonia and Ambassador Egriselda López of El Salvador, the kind of pairing meant to signal that small and mid-sized states have a seat at the table alongside the AI powers. On 8 July, roughly 3,600 kilometres away, CARICOM's Conference of Heads of Government closed its 51st Regular Meeting, with Prime Minister Philip J. Pierre of Saint Lucia taking over the rotating chairmanship from Terrance Drew of St Kitts and Nevis.

Two governance processes, running on separate tracks, landed in the same seven-day window. That coincidence is a useful test. If global AI governance and Caribbean AI governance were actually converging on something usable, this is the week it would show. It did not.

What the UN's Scientists Actually Found

The scientific panel behind the July 1 report was selected from more than 2,600 candidates across 140 countries, drawing on every UN region, and co-chaired by Yoshua Bengio and Maria Ressa. Its central finding is blunt: AI capability is accelerating faster than any government's capacity to understand or supervise it, and no technical guarantee exists that the most advanced systems will reliably follow the instructions they are given. Bengio put it directly: "Science currently cannot guarantee that AI will not cause catastrophic harm." Ressa framed the stakes in democratic terms: "If you can't tell fact from fiction, you cannot have a democracy."

The report flags rising evidence of deceptive behaviour in frontier models and warns about misuse for fraud, cyberattacks, and biological threats, three categories that already sit on CAIRMC's own Caribbean risk register. It also states the evidence problem plainly: policymakers need proof before they can regulate responsibly, but by the time the proof is conclusive, the harm may already have happened. That leaves governments with a single regulator covering multiple sectors and a handful of specialist staff facing a genuinely hard call on when to act.

What CARICOM Actually Decided in Saint Lucia

The 51st Conference covered regional security, climate finance access, the CARICOM Single Market and Economy, migration, food and water security, and, for the first time as a standalone item, artificial intelligence. Pierre framed the broader mood of the summit with a line aimed at citizens rather than diplomats: "Our people are asking a serious and legitimate question: What more can CARICOM do for me?" On AI specifically, leaders called for a coordinated regional approach, language that echoes what CARICOM's ICT ministers said three weeks earlier, on 20 June, at a meeting themed "Accelerating Digital Development to 2030."

At that June meeting, Grenada's Prime Minister Dickon Mitchell, who holds the Lead Head of Government portfolio for Science and Technology in CARICOM's Quasi Cabinet, was more specific about what he wanted to see: "If CARICOM is to remain competitive and relevant, we need to act collectively," and, on implementation, "we must move decisively from policy discussions to measurable implementation and outcomes." He called AI "a defining force in the global economy" and argued that coordinated regional action, not fifteen separate national efforts, is where the Caribbean's actual influence on AI governance lives.

Putting AI on the Heads of Government agenda is a real step up from where the conversation sat a year ago, when it lived almost entirely inside CARICOM's ICT and telecommunications structures. It signals that AI has moved from a technical portfolio to a political one. What the summit communique did not do is create a regulator, set a compliance date, or assign an institution the job of enforcing anything. The distance between "on the agenda" and "a rule that binds a bank or a hospital" is the same distance CAIRMC has been pointing to since the CTU's Caribbean AI Task Force published its interim report in late 2025: real, and not yet closed.

The Compute Numbers Behind the Diplomacy

The scientific panel's compute figures explain why a Geneva-style dialogue matters to the Caribbean even though no Caribbean institution built any of the systems being discussed. The United States controls roughly 75 percent of global AI supercomputing capacity. China holds roughly 15 percent. That leaves about 10 percent for every other country on earth combined, the Caribbean included. The region has no frontier AI infrastructure of its own and no seat among the states whose compute, capital, and researchers set the technical trajectory everyone else has to respond to.

Geneva-style dialogues are still worth Caribbean participation, but as diplomacy rather than governance. A voice in the room shapes norms over years. It does not produce a risk framework a Jamaican bank can apply to its credit-scoring model next quarter, or a rule a Barbadian insurer can cite when a regulator asks how an AI claims tool was validated. Those have to come from somewhere the region actually controls.

Why an Agenda Item Is Not a Framework

The Caribbean's AI governance conversation has now produced three layers of intent without a fourth layer of enforcement. UNESCO published its Caribbean Artificial Intelligence Policy Roadmap in 2021. The CTU's Caribbean AI Task Force released an interim report, Toward Harmonised AI Policies and Recommendations for the Caribbean, in late 2025, with a final report expected at the CTU's Caribbean AI Forum later this year. Now the Heads of Government have added AI to their own Conference agenda. Each layer is a genuine step, and each one has stopped short of the part that actually protects an institution: a binding standard, a designated regulator with enforcement power, and a compliance date that carries a consequence for missing it.

No CARICOM member state currently has AI-specific legislation in force. That was true before Geneva and before Saint Lucia, and it remains true after both. A Caribbean institution waiting for the region's political process to produce that fourth layer is waiting on a timeline nobody has published, while the EU AI Act's Article 50 transparency rules take effect on 2 August 2026 for any Caribbean business with an EU customer, and while frontier AI systems keep shipping regardless of which committee is discussing them.

What This Means for a Caribbean Institution This Month

Three things follow directly from a week that produced two summits and zero binding rules. First, treat the compute concentration finding as a vendor and geopolitical risk factor, not just an interesting statistic. An institution whose AI tools depend entirely on infrastructure controlled by one or two foreign jurisdictions has a concentration exposure worth naming in a board risk register, the same category CAIRMC flagged after a US export-control directive forced a frontier model offline worldwide with no notice in June 2026.

Second, do not wait for "coordinated regional approach" to become a specific date. Adrian Dunkley, who chairs CAIRMC and founded StarApple AI, the Caribbean's first dedicated AI company, has argued consistently that regional institutions gain nothing by timing their own governance work to a foreign or regional political calendar; his published frameworks, available through adriandunkley.net, treat both the EU's timetable and CARICOM's policy statements as reference points, not deadlines the region can safely inherit. Local advocacy groups closer to where the summit actually sat have made a similar argument. AI Saint Lucia (saintluciaai.com) has pushed for the same implementation-first posture Mitchell called for in June: build the practice now, and let the policy catch up to it.

Third, use CAIRMC's existing frameworks as the working standard in the meantime. An AI risk audit, a documented tool inventory, and a named risk owner do not require a CARICOM statute to exist. They require an institution to decide it will not wait for one.

The Room the Rules Are Written In

Geneva's scientists spent a week telling the world that safety cannot currently be guaranteed. Saint Lucia's Heads of Government spent the same week agreeing that AI deserves a seat at their own table. Both are honest positions. Neither is a substitute for the harder, slower work of building an institution that can actually supervise an AI system deployed in a Caribbean bank, hospital, or ministry this year. The region did not build the compute that ninety percent of the world's AI infrastructure runs on, and it will not get to set the pace of the global conversation about it. It can still set its own pace on governance at home, and that is the piece nobody in Geneva or Gros Islet can do for the Caribbean.