ACCRA, Ghana: A warning from a senior safety researcher at the artificial intelligence company Anthropic has triggered new international debate about whether more powerful AI systems could eventually escape human control. Evan Hubinger, who works on AI alignment at Anthropic, said he personally believes there is a greater than 10% chance that advanced AI could cause human extinction within the next decade. His comments followed the resignation of researcher Jacob Coxon, who previously worked at both Anthropic and OpenAI. Coxon accused the companies of moving too quickly toward developing “self-improving superintelligence” without adequate safeguards. Anthropic develops Claude, an AI assistant used for writing, research, data analysis, coding, and other tasks. The warnings are serious, but Ghanaians should understand what these claims mean and what they do not mean.
The 10% figure is not a proven scientific forecast.
No universally accepted scientific formula can calculate the exact probability that AI will cause human extinction. Hubinger’s number should be seen as his informed personal opinion, based on his experience with AI development and safety research. It does not prove that humanity has a definite 10% chance of being destroyed by AI. Computer scientist Geoffrey Hinton, a major figure in AI research, has also said that a 10% estimate might not be unreasonable. However, he admitted that no one knows how to calculate this probability reliably, since humanity has never faced technology like this before. Business Insider. This prediction should be taken seriously as a warning from experts, but it should not be reported as if it is certain to happen.
Current AI is not an independent superpower.
Today’s AI systems can write articles, analyze data, translate languages, create images, help programmers, and answer questions. They can do some tasks faster than humans, but they still make mistakes, sometimes give wrong information, and rely on computer systems built by people. Current chatbots do not independently control governments, nuclear weapons, banks, electricity networks, or military forces. They cannot automatically take over society simply because someone asks them a question. For an AI system to cause the kind of destruction researchers describe, several safety measures would probably have to fail. A future AI would need to become much more capable and independent, gain access to critical infrastructure or dangerous resources, break through cybersecurity protections, and resist efforts to shut it down. Governments, companies, or individuals would also have to give such a system excessive authority without effective supervision. This scenario is plausible enough to justify serious preparation, but there is no evidence that today’s regular chatbots are about to destroy humanity.
What is AI alignment?
AI alignment means making certain an AI system keeps following human goals, values, and safety rules as it becomes more advanced. The challenge is that human instructions can be partial or misunderstood. For example, telling an AI to “reduce crime by every available means” would not automatically make it clear that it must also protect privacy, human rights, and the presumption of innocence. A very advanced system could, in theory, find an unexpected or harmful way to reach a goal that is not clearly defined. That’s why researchers are working to make sure AI follows not just the exact words of instructions, but also the values and limits behind them. Hubinger said the risk from today’s AI models is still low. He is mainly worried about possible future “superintelligent” systems and whether people will have good ways to control them.
Immediate AI risks deserve attention in Ghana.
While human extinction is still a worst-case scenario, AI is already creating real challenges Ghana needs to address. Fraudsters can use AI-generated voices, images, and videos to impersonate public officials, relatives, business leaders, and celebrities. Deepfake content can spread political falsehoods or damage a person’s reputation. AI-generated messages can also make online scams appear more professional and believable. Other concerns comprise privacy violations, biased automated decisions, academic dishonesty, inaccurate health information, and the possible displacement of workers whose duties can be automated. These current risks are more urgent for Ghana than the idea of AI taking over the world on its own. Ghanaians should verify suspicious videos and voice recordings, especially when they request money or personal information. They should also check important AI-generated information about health, law, employment, or finance with qualified professionals.
AI also delivers considerable opportunities.
AI should not be seen only as a threat. When applied responsibly, it can help with education, healthcare, farming, journalism, public services, and business growth. Farmers could use AI to get weather, crop, and market updates in local languages. Teachers could make learning materials for students with different needs. Health workers could use well-tested systems to analyze information and improve their work, while businesses could automate standard tasks and serve customers better. AI can also communicate in languages like Twi, Ewe, Ga, and Dagbani, though the quality can differ. This helps make digital services easier to use for people who are not comfortable with English. The aim should be to help Ghana use AI in useful ways while protecting people from misinformation, discrimination, privacy problems, and fraud.
Governments and technology companies must act.
Technology companies should subject powerful AI models to independent safety testing before deployment. They should restrict access to dangerous capabilities and require human approval for high-impact actions. Companies must also strengthen cybersecurity, monitor system use, report serious incidents, and maintain reliable procedures to isolate or turn off unsafe systems. Governments should work together internationally because AI systems and cybersecurity risks cross borders. Shared safety standards could require testing, incident reporting, and stronger controls for the most powerful models, without making low-risk applications too hard to use. Ghana must also develop its own AI governance framework, invest in digital education, and ensure that local experts engage in global discussions. The country should not depend entirely on regulations and technologies designed elsewhere.
Concern should lead to preparation, not panic.
The warnings from Anthropic researchers should not be dismissed. People working closely with advanced AI systems may see emerging capabilities that deserve public and government attention. At the same time, scary predictions should be shared honestly. The claim of more than a 10% extinction risk is an expert’s opinion in a situation of great uncertainty, not a confirmed scientific fact. Ghanaians do not need to stop using AI. They should use it thoughtfully, verify important information, protect personal data, and keep humans responsible for decisions affecting health, employment, finance, justice, and public safety. AI can bring great benefits, but we should pursue them with safeguards, transparency, and accountability. The right response is not blind trust or panic, but knowledgeable and accountable action.
By Michael Abisa
September 14, 2026
