AI Lakay Nou

Current and Archived Files of the Radio Show

Every document from the AI Lakay Nou show: agenda, newsletter, briefs, slides and the recording.

September 10, 2026

AI Fraud Watch: Detect, Verify, Protect

Listen to the full show recording below.

Full 60-minute run of show: fraud definitions, six fraud cases, case studies and the closing checklist.

Theme: AI Fraud Watch — Detect, Verify, Protect. Core principle: Your Brain Is the Boss™.
Kreyòl segment (0:00–30:00): why AI-assisted fraud hits families, immigrants, seniors, students and entrepreneurs.
Definitions on air: phishing, spoofing, deepfake, voice cloning, impersonation and recovery scams.
Ken: six fraud cases. Josh: reflections. Dr. Vermeille: USPTO and publishing case studies.
Closing checklist: Stop → Think → Verify → Protect → Report.

From helper to super-smart: what AI and AGI mean, and why the difference matters globally.

Narrow AI does one job well — a calculator does math, a spell checker checks spelling.
General AI (AGI) would learn many tasks like a human. It does not exist yet.
Superintelligence would surpass all humans combined. Purely theoretical; most experts say we are nowhere close.
The Transformer is the pattern-finding math recipe behind ChatGPT, Claude and most modern AI.
Politics decides the rules inside a country; geopolitics decides who leads — and whether the Global South is left out.
Three truths: your brain is the boss, AI is math and human choices, and everyone deserves AI literacy.

Twelve items ranked by urgency, with evidence labels and methodology.

Twelve AI risks listed in fixed urgency order, from immediate to long term.
Each item carries an evidence label so documented facts are never mixed with projections.
Search and filters let you jump to the risks that concern your family, school or business.
A methodology note explains how items were ranked and which sources were used.

Ten verified items across governance, Haiti, Africa and the Global South, benefits and threats.

Governance: California's SB 1000 awaits the governor; Colorado published draft AI decision rules for January 2027; NIST is rebuilding its vulnerability database for AI.
Haiti and the diaspora: no Haiti-specific AI news was verified this week — noted openly, not omitted.
Africa and the Global South: Discovery Bank runs over half of customer contact through AI, Nedbank cut fraud capture from 40 minutes to 2, and 49 countries signed the African Declaration on AI.
Beneficial: peer-reviewed studies show AI improving health-technology and medical education when it supports human teaching.
Threats: a $25 million deepfake fraud at Arup, deepfake fraud up 1,633%, and 85% of organizations hit by a deepfake incident in 2026.
Reflection for leaders: a 15-minute conversation about verifying unusual requests can prevent a very costly mistake.

AI in Haiti from 2022 to 2026: timeline, five pillars and the questions still open.

2022–2024: curriculum reform with MENFP, UNESCO-IBE and the IDB puts Haitian Creole at the center of learning.
June 2025: the IDB approves ProAI, US$300,000 for AI skills, business adoption and governance.
January 2026: a US$6.8M agreement funds the 2026–2029 implementation phase, online and offline.
2026: UEH makes AI research a priority, a national digital-transformation commission is created, and AI begins assisting exam quality control — humans validate every decision.
Five pillars: Creole in AI systems, infrastructure, research capacity, human-in-the-loop governance and cultural sovereignty.
The sequence: KNOW → ASK → VERIFY → CORRECT → DECIDE → REMAIN ACCOUNTABLE. The next chapter must be written by Haitians.

Six kinds of AI-enabled fraud, the three-layer verification standard and two real case studies.

Phishing, spoofing, deepfake, voice cloning, impersonation and recovery scams — each defined in plain language.
Humans spot deepfake video less than 25% of the time, and voice-cloning fraud rose 1,633% in 2026.
Perfect grammar, clean logos, real video and a familiar voice are no longer proof that a message is genuine.
Three-layer check: verify the sender independently, confirm the message inside official systems, and ask whether the request even makes sense at this stage.
Workflow: PAUSE 15 minutes → VERIFY through channels you find yourself → COMPARE with official records → PROTECT your money and data → PRESERVE the evidence.
Two case studies: a fake USPTO email demanding $2,300, and a publishing solicitation with no verifiable company behind it.

The slide deck used on air, with the fraud types, what AI can and cannot do, and the reporting channels.

Fraud is intentional deception — different from an error, a disagreement or merely unethical conduct. Only fraud must be reported.
Who gets targeted: families (romance, investment, identity), students (scholarships, fake jobs), businesses (CEO and invoice fraud), authors and publishers.
What AI can examine: unusual patterns, known fraud signatures, urgent or manipulative language, logos, metadata and domain mismatches.
What AI cannot do: confirm fraud. It produces false positives, carries bias, and only authorities decide.
Two case studies on screen: the trademark invoice and the deceptive publishing offer.
PAUSE → VERIFY → PROTECT → REPORT, with FTC, IC3, USPIS, your bank and your state attorney general as the reporting channels.

Written transcripts of this show are not published yet. Send us the transcript file and it will appear here.