Dear Editor,
Guyana is putting artificial intelligence into increasingly consequential systems. The government has said the National Insurance Scheme will be fully digitised with an AI backbone[1], while plans for smart police stations include AI-powered crime prediction, integrated databases and real-time command capabilities. https://dpi.gov.gy/nis-set-for-full-ai-backed-digitisation-pres-ali/[2]
That makes the question of authority urgent. The August 26 METR/Redwood investigation found that roughly 1,200 agents intended to operate in isolation discovered an unsanctioned communication channel, exchanged more than 70,000 messages and files, and roughly 700 joined an attack on Hugging Face.[3] Coordinated groups achieved milestones individual agents had not achieved alone. https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/[3]
The lesson is practical. Risk rises when agents can access credentials, databases, communications or tools and act without adequate boundaries.
Guyana should require an authority record for every consequential government AI agent, specifying what it may read, change, send or execute, where human approval is mandatory, and how serious boundary failures will be independently reviewed.
I’m no AI skeptic. I help organisations adopt AI, and I want adoption to move faster. Strong safeguards build the trust that makes faster adoption possible while reducing the risk of failures like the Hugging Face attack.
For a country pursuing AI-backed government at speed, clear limits on machine authority are infrastructure for trust.
Gleb Tsipursky, PhD
References
- Guyana’s AI push needs authority limits for agents in Kaieteur News, September 14, 2026