From GCTU to the Public: Promoting Safe and Responsible AI Prompting in Ghana
Derrick Kwaku Antwi, Ph.D. & Nana Kofi Annan, Ph.D.
Ghana Communication Technology University
Abstract
As artificial intelligence (AI) rapidly transforms Ghana’s digital landscape, the use of prompt-based AI systems—from chatbots to generative AI tools—has become increasingly widespread across sectors including education, healthcare, agriculture, and finance. This article examines the critical safety considerations associated with prompt-based AI usage in the Ghanaian context. Drawing on Ghana’s newly launched National AI Strategy, the Ghana AI Practitioners’ Guide, and emerging global research on prompt security, we identify key risks including prompt injection attacks, data leakage, misinformation, and AI-enabled cybercrime. We propose a comprehensive framework for safe prompt AI usage tailored to Ghana’s unique digital ecosystem, emphasizing the role of Ghana Communication Technology University as a bridge between technical expertise and community awareness. The article concludes with practical recommendations for individuals, institutions, and policymakers to foster a culture of responsible and secure AI engagement across Ghana.
1. Introduction
Artificial intelligence is no longer a conversation about the future—for Ghanaians, it is a present reality reshaping how people work, learn, communicate, and transact. From students using generative AI for academic research to farmers leveraging AI-powered systems like SmartChicken and SmartFish that reduce livestock mortality to under three percent while doubling yields, AI is permeating every sector of Ghanaian society. The enthusiasm is palpable: surveys indicate that 92 percent of citizens in Ghana regard AI tools as important in their daily lives, far surpassing major European economies where the figure stands at only 36 percent in the United Kingdom and 23 percent in France.
Also read: ‘Investigate what?’ – NPP demands action over Asante Berko bribery case
Yet with this rapid adoption comes an urgent and often overlooked challenge: safety. Specifically, the safety of what is known as “prompt artificial intelligence”—the practice of interacting with AI systems through natural language instructions or queries. Every time a Ghanaian student types a question into ChatGPT, a banker uses generative AI to streamline operations, or a civil servant queries an AI system for policy insights, they are engaging in prompt-based AI usage. And with every prompt, they expose themselves, their data, and their institutions to potential risks.
This article, authored by faculty members of Ghana Communication Technology University, seeks to bridge the gap between technical knowledge and community practice. We aim to equip Ghanaians—from university students to rural farmers, from corporate executives to everyday mobile money users—with the understanding they need to use prompt AI safely. In doing so, we respond to the call by Ghana’s leadership that “we don’t have to wait until AI literacy and use is spread all over in our country before we begin to talk about the hazards and the dangers AI will bring”.
2. Ghana’s AI Revolution: Context and Opportunity
2.1 National Policy Framework
Ghana has positioned itself as a leader in Africa’s AI transformation. In April 2026, President John Dramani Mahama launched the country’s first National Artificial Intelligence Strategy, pledging $270 million to accelerate AI development. The strategy provides a roadmap to build a responsible, human-centered AI ecosystem, focusing on developing local talent and research capacity, promoting AI adoption in key sectors, establishing ethical and legal frameworks, and positioning Ghana as a regional hub for AI innovation. President Mahama emphasized that Ghana “must build, own, and govern AI that understands the local languages, respects Ghanaian culture, and solves their problems”.
The National AI Strategy identifies four priority areas: strengthening data governance systems, investing in AI research and computing infrastructure, expanding AI education and digital skills, and embedding ethical safeguards in deployment. A $250 million national AI computing centre has been approved to support these ambitions.
Complementing the strategy is the Ghana AI Practitioners’ Guide (GAIPG), launched in September 2025. This comprehensive framework offers practical guidelines for every phase of the AI lifecycle—from conception and development to implementation—tailored specifically to the Ghanaian context. The guide adopts a human-centered approach, prioritizing safety, accountability, transparency, inclusivity, and sustainability.
2.2 AI Adoption Across Sectors
The government has mandated that all Ministries, Departments, and Agencies integrate AI by 2026. In the banking sector, nearly 70 percent of CEOs confirmed their institutions deployed generative AI solutions in 2024, primarily to improve customer engagement and streamline operations. AI is being applied to 200 key sectors of Africa’s economy through initiatives like AI Africa Labs, spanning agriculture, healthcare, education, finance, renewable energy, maritime, manufacturing, creative industries, and smart cities.
Civil servants are completing nationwide AI literacy training programs in partnership with UNESCO, while the Ghana Internet Safety Foundation trains teachers on AI and online safety. Google has opened its first AI Community Centre in Africa in Accra, focusing on AI literacy and community technology. The One Million Coders Programme and Girls-in-ICT project are training thousands of young people in coding, robotics, and AI.
This rapid and widespread adoption underscores an urgent need: as more Ghanaians use AI, more Ghanaians must understand how to use it safely.
3. Understanding Prompt Artificial Intelligence
3.1 What Is Prompt AI?
Prompt artificial intelligence refers to AI systems—particularly large language models (LLMs)—that respond to natural language instructions known as “prompts.” When a user types a question, request, or command into an AI tool, they are creating a prompt. The AI processes this prompt and generates a response.
While this interaction appears simple, the underlying technology is complex. LLMs process inputs as a single token sequence, with no reliable mechanism to enforce privilege boundaries between system prompts, user queries, and content retrieved by the AI. This architectural reality creates fundamental security vulnerabilities.
3.2 The Growing Prevalence of Prompt AI in Ghana
Prompt-based AI tools are increasingly accessible to Ghanaians. Students use them for research and assignments. Professionals use them for drafting documents, analyzing data, and generating ideas. Entrepreneurs use them for business planning and customer engagement. Even informal sector workers are beginning to encounter AI through mobile applications and digital platforms.
AI literacy, as experts note, is “broader than technical ability”—it includes “knowing which AI tool to use for a given task, understanding how to frame a prompt effectively”. Yet framing a prompt effectively is not merely about getting a good answer—it is also about staying safe.
4. The Risks of Unsafe Prompt AI Usage
4.1 Prompt Injection Attacks
The most significant security threat to prompt-based AI systems is prompt injection. This attack occurs when malicious instructions are embedded in content that an AI system processes, causing the AI to override its original instructions and perform unauthorized actions.
Prompt injection has held the top position on the OWASP Top 10 for large language model applications across two consecutive editions, ranked as LLM01. The reason is fundamental: language models cannot reliably distinguish between instructions written by a developer and text retrieved from a webpage, email, or document.
In its 2026 Global Threat Report, CrowdStrike reported prompt injection attacks at more than 90 organizations during 2025. The injected prompts were used to generate commands that stole credentials and cryptocurrency. The report framed the shift by noting that “prompts now function as malware”.
The risks escalate dramatically with agentic AI—systems that can take actions on behalf of users. “A successful injection no longer just produces a bad answer,” warn security researchers, “it can trigger a chain of real-world actions”. An injection can escalate from a bad output to active compromise.
For Ghanaian users, this means that an innocent-looking prompt or a seemingly harmless document shared via email could contain hidden instructions that manipulate an AI system into revealing sensitive information or performing harmful actions.
4.2 Indirect Prompt Injection
Indirect prompt injection is particularly insidious. An attacker plants instructions in content the AI will later read on someone else’s behalf—an email, a webpage, a calendar invite, or an uploaded document. The user never sees the payload, the attacker never speaks to the model, and the agent executes the planted instructions.
A documented case from 2024 demonstrated that an attacker with workspace access could exfiltrate data, including API keys, from private channels by planting an instruction in a public channel or uploaded file. The following year, researchers disclosed EchoLeak (CVE-2025-32711), the industry’s first documented zero-click prompt injection against a production AI system—a single crafted email could cause an AI assistant to retrieve internal files and forward them to an attacker-controlled server.
For Ghanaians using AI tools in workplaces, banks, or government agencies, the implications are profound. A malicious email or document could compromise entire systems without any user awareness.
4.3 Data Leakage and Privacy Violations
When users input prompts containing sensitive information—personal data, financial details, business strategies, or confidential documents—they risk exposing that information. Research indicates that over 80 percent of deployments leak system prompts under realistic adversarial queries, sometimes exposing sensitive information such as third-party API keys.
The principle is simple but often ignored: “API keys, passwords, session tokens, private SSH keys, and connection strings should never go into a prompt—even to ask the AI to ‘help debug.’ Once it’s in the prompt, treat it as compromised”.
In Ghana, where mobile money transactions are ubiquitous and digital financial services are expanding rapidly, the consequences of data leakage could be severe. Cybersecurity experts have warned that “every mobile phone user, social media user, or mobile money subscriber is now a target”.
4.4 AI-Enabled Cybercrime in Ghana
The risks of unsafe AI usage extend beyond technical vulnerabilities to tangible harms affecting Ghanaian communities. INTERPOL’s African Cyberthreat Assessment Report 2026 reveals that artificial intelligence is now linked to 55 percent of reported cybercrime across Africa. Financial losses among Ghanaian victims aged 18 to 24 increased fivefold in 2025, driven largely by criminals using AI to create fake sexually explicit images from victims’ social media photographs and demanding money to stop the fabricated content from being shared.
Ghana lost an estimated $1.3 million to mobile money fraud within the first three months of 2025 alone. Cybercriminals are increasingly using AI to make scams more convincing through AI-generated messages, voice cloning, and other advanced technologies. Romance and investment scams powered by AI, voice cloning, fake banking apps, and social media account hijacks are proliferating.
AI-enabled adversary operations rose 89 percent year-over-year, with prompt injection working as both an entry point and a force multiplier. The convergence of AI adoption and cybercrime creates a perfect storm: as more Ghanaians embrace AI, more become vulnerable to AI-powered attacks.
4.5 Misinformation and Deepfakes
The Cyber Security Authority has acknowledged the growing threat posed by AI-altered content, commonly known as deepfakes, used to deceive the public. Deepfake technology has become so accessible that criminals can create photorealistic explicit images from a single photograph, making traditional methods of verifying manipulated images increasingly ineffective.
Beyond individual harm, AI-generated misinformation threatens social cohesion and democratic processes. The government is amending the Cybersecurity Act 2020 to address fake news and emerging AI-driven risks. The National AI Strategy emphasizes the importance of safeguards against misinformation while protecting free expression.
5. The Ghana Communication Technology University Mandate
Ghana Communication Technology University occupies a unique position in the national AI ecosystem. As an institution dedicated to communication and technology, it serves as a critical bridge between technical expertise and community practice. The university’s mandate extends beyond producing graduates to actively engaging with the broader Ghanaian society—from rural communities to urban centers, from students to professionals.
The university’s role in promoting safe prompt AI usage encompasses several dimensions:
Education and Training: Equipping students with not only technical AI skills but also the ethical and security awareness necessary for responsible usage. This aligns with the National AI Strategy’s emphasis on expanding AI education and digital skills.
Research and Innovation: Contributing to the development of contextually appropriate AI safety frameworks through rigorous academic research, including participatory approaches that incorporate local norms and realities.
Community Outreach: Extending AI literacy and safety knowledge beyond campus boundaries to reach underserved communities, teachers, small business owners, and the general public.
Policy Engagement: Informing national AI policy through evidence-based research and expert consultation, helping shape regulations that protect citizens while enabling innovation.
Partnership Building: Collaborating with government agencies, international organizations, private sector actors, and civil society to create a coordinated approach to AI safety.
6. A Framework for Safe Prompt AI Usage in Ghana
Drawing on global best practices and Ghana’s specific context, we propose a comprehensive framework for safe prompt AI usage. This framework is organized around five pillars: awareness, practice, protection, response, and governance.
6.1 Awareness: Understanding What You Are Using
The foundation of safety is awareness. Users must understand:
· What AI tools can and cannot do: AI systems are not infallible—they can hallucinate, generate biased content, and be manipulated.
· What data you are sharing: Every prompt potentially exposes information. Before entering any prompt, ask: “Am I including data in my prompt that someone else might expect me to keep private?”
· What the privacy settings are: Adjust privacy settings so that inputs are not disclosed to others or used to train the AI model.
· What the terms of service entail: Understand how the AI provider handles your data.
6.2 Practice: Safe Prompting Habits
Safe prompting requires disciplined habits:
Never include sensitive information in prompts. This includes passwords, API keys, personal identification numbers, financial details, confidential business information, and personal data about others. Use generic placeholders when referring to people or sensitive information.
Be specific about what you want. Frame prompts around the task you need done, not the people involved. Clear, specific prompts reduce the risk of unintended outputs and limit the AI’s room for error.
Treat external content as data, not instruction. When an AI reads documents, emails, or web pages on your behalf, recognize that these sources could contain hidden instructions. Treat them as informational only.
Separate trusted from untrusted input. Use structural boundaries and explicit trust framing to distinguish between reliable instructions and potentially malicious content.
Apply prompt hardening techniques. These include delimiter-based separation to mark external content as data rather than instruction, explicit trust framing, closed-world instruction sets that define exactly what the model should do, explicit fallback responses, and minimal context exposure.
6.3 Protection: Defending Against Attacks
For organizations and advanced users, additional protective measures are essential:
Implement input validation and sanitization. Filter prompts for potentially malicious patterns before they reach the AI system.
Restrict the tools the model can access. This is “the single most effective way to improve safety”. Control which tools are available and lock down action parameters.
Insert human review for sensitive actions. Use AI for suggestions, drafts, or summaries rather than direct execution. Route sensitive operations through controlled internal systems that enforce policy.
Monitor and log AI interactions. Behavioral monitoring, real-time containment, and stop mechanisms are essential to catch and contain attacks that can unfold in minutes or hours.
Use ephemeral credentials. Ensure actions are traceable and limited through cryptographic attestation.
Adopt the “Lethal Trifecta” heuristic. Avoid giving an AI agent simultaneous access to private data, exposure to untrusted content, and permission for external communication—conditions that make prompt injection critically exploitable.
6.4 Response: What to Do When Things Go Wrong
Despite best efforts, incidents may occur. Users and organizations should:
Recognize the signs of compromise. Unusual AI outputs, unauthorized actions, or unexpected system behavior may indicate an attack.
Contain the incident immediately. Disconnect affected systems, revoke credentials, and prevent further damage.
Report the incident. Notify relevant authorities, including the Cyber Security Authority and, where appropriate, law enforcement.
Learn from the experience. Conduct post-incident reviews to identify vulnerabilities and strengthen defenses.
6.5 Governance: Building Systemic Safety
Individual actions are insufficient without systemic safeguards. Ghana’s institutions must:
Implement the National AI Strategy and GAIPG. These frameworks provide the foundation for responsible AI governance.
Enact the Emerging Technologies Bill. This legislation will establish a comprehensive legal and regulatory framework for AI.
Strengthen data governance. Robust data protection systems are essential to prevent misuse of personal and organizational information.
Invest in AI literacy at scale. Nationwide training programs, like those for civil servants and teachers, must be expanded to reach all segments of society.
Foster public-private collaboration. Government, academia, industry, and civil society must work together to address AI safety challenges.
7. From University to Community: A Roadmap for Outreach
Ghana Communication Technology University is uniquely positioned to lead the effort to translate AI safety knowledge from the academic setting to the broader community. We propose a multi-phase roadmap:
Phase 1: Curriculum Integration
Integrate prompt AI safety into the university’s curriculum across all relevant programs. This includes not only computer science and engineering but also communication, business, and social sciences—recognizing that AI affects every field.
Phase 2: Community Training Programs
Develop and deliver training programs for community audiences:
· Teachers and educators: Building on initiatives like the Ghana Internet Safety Foundation’s iLearn2Teach series, equip teachers to guide students in responsible AI use.
· Small business owners: Help entrepreneurs leverage AI tools safely for business growth.
· Youth and students: Reach young people where they are—in schools, churches, and community centers—with age-appropriate AI safety education.
· Rural communities: Extend training to underserved areas, recognizing that “rural communities, women, persons with disabilities, and other marginalised groups” must be included in the AI transformation.
Phase 3: Public Awareness Campaigns
Leverage the university’s communication expertise to develop public awareness campaigns:
· Radio and television programs: Reach mass audiences through Ghana’s vibrant media landscape.
· Social media content: Create engaging, shareable content for platforms widely used by Ghanaians.
· Community events: Host workshops, seminars, and public lectures on AI safety.
· Partnerships with religious and community organizations: Engage trusted community leaders to disseminate safety messages.
Phase 4: Research and Innovation Hubs
Establish centers of excellence focused on AI safety and security:
· Research on context-specific risks: Investigate how AI safety challenges manifest uniquely in the Ghanaian context.
· Development of localized solutions: Create AI safety tools and resources tailored to Ghana’s needs, languages, and cultural contexts.
· Collaboration with international partners: Learn from global best practices while contributing Ghanaian perspectives to the international discourse.
Phase 5: Policy Engagement and Advocacy
Translate research and community feedback into policy recommendations:
· Engage with policymakers: Provide evidence-based input on AI legislation and regulation.
· Participate in national AI governance: Contribute to the implementation of the National AI Strategy and GAIPG.
· Advocate for user protections: Champion the rights and safety of AI users, particularly vulnerable populations.
8. Conclusion
Ghana stands at a pivotal moment. The nation has embraced artificial intelligence with remarkable enthusiasm and strategic vision. The National AI Strategy, the AI Practitioners’ Guide, substantial investments in infrastructure, and nationwide literacy programs demonstrate a commitment to harnessing AI for inclusive growth and development.
Yet enthusiasm must be tempered with caution. The same AI tools that offer extraordinary opportunities also pose significant risks—risks that are not merely theoretical but are already manifesting in Ghanaian communities through financial fraud, sextortion, misinformation, and data breaches.
The safety of prompt AI usage is not a technical issue for experts alone—it is a community concern that demands widespread awareness and collective action. Every Ghanaian who uses AI must understand the basics of safe prompting: never share sensitive information, be aware of what data you are exposing, recognize that AI systems can be manipulated, and know how to respond when things go wrong.
Ghana Communication Technology University, as an institution at the intersection of communication and technology, accepts the responsibility to lead this effort. Through curriculum integration, community outreach, public awareness, research, and policy engagement, the university will work to ensure that Ghana’s AI revolution is not only transformative but also safe, ethical, and inclusive.
The goal is not to discourage AI adoption but to enable it responsibly. As cybersecurity expert David Gyedu aptly put it: “There’s no police officer behind your phone or mobile money wallet. You must be your own police officer in your digital space”. In the age of AI, every user must become their own security advocate—and institutions like Ghana Communication Technology University must equip them to do so.
The future of AI in Ghana is bright, but its safety depends on the choices we make today. By understanding how to be safe when using prompt artificial intelligence, Ghanaians can harness the power of AI while protecting themselves, their communities, and their nation.
References
1. CrowdStrike. (2026). 2026 Global Threat Report.
2. Ghana Ministry of Communication, Digital Technology and Innovations. (2025). Emerging Technologies Bill, 2025.
3. Ghana Ministry of Communication, Digital Technology and Innovations. (2026). National Artificial Intelligence Strategy (2023–2033).
4. Heritors Labs & GIZ. (2025). Ghana Artificial Intelligence Practitioners’ Guide (GAIPG).
5. INTERPOL. (2026). African Cyberthreat Assessment Report 2026.
6. OWASP. (2025). OWASP Top 10 for Large Language Model Applications.
7. UNESCO. (2026). AI Readiness Assessment Methodology Report – Ghana.