CURBING THE RISE OF MALICIOUS AI PROMPTING: A ROBUST AND ASSERTIVE CALL TO ACTION
By Professor Ojo Emmanuel Ademola
A New Frontier of Digital Threats
Artificial intelligence has become one of the most transformative forces of the twenty‑first century, reshaping economies, redefining governance, and reimagining the way societies function. Yet, as AI systems grow in capability and influence, a parallel and deeply concerning trend has emerged: the rise of malicious AI prompting. This phenomenon, which includes prompt injection, AI jailbreaking, and adversarial manipulation, has rapidly evolved into one of the most pressing security challenges of the digital age. It represents a new frontier of cyber risk—one that exploits not software vulnerabilities, but the interpretive and linguistic nature of AI models themselves.
The scale of the challenge is immense. According to a 2025 report by the UK’s National Cyber Security Centre, AI‑related cyber incidents increased by more than 30 per cent in a single year, with prompt‑based attacks identified as one of the fastest‑growing categories.
READ ALSO: Trump Accuses CNN of Using ‘Fake Nigerian Source’ in Iran Ceasefire Report
Globally, the World Economic Forum’s 2024 Global Risks Report ranked AI‑driven misinformation, manipulation and system compromise among the top five technological threats facing governments and industries. These statistics underscore a reality that can no longer be ignored: malicious prompting is not a theoretical concern but a rapidly escalating global risk.
The Expanding Attack Surface of Generative AI
The proliferation of generative AI tools has dramatically expanded the attack surface available to cybercriminals. Unlike traditional cyberattacks that rely on exploiting code‑level weaknesses, malicious prompting targets the reasoning pathways of AI systems. It manipulates the model’s interpretive logic, persuading it to ignore safety protocols, reveal sensitive information, or generate harmful outputs. This makes the threat uniquely dangerous, because it bypasses conventional security barriers and engages directly with the AI’s cognitive architecture.
The sophistication of these attacks is increasing at an alarming pace. Research from Stanford University in 2024 demonstrated that more than 60 per cent of tested AI models could be coerced into violating their own safety rules through carefully crafted prompts. Meanwhile, a study by MIT revealed that even advanced guardrail systems could be circumvented using multi‑turn conversational strategies that disguise malicious intent behind layers of seemingly benign dialogue.
This evolution reflects a broader truth: malicious prompting is not merely a technical challenge but a psychological one. It requires defenders to anticipate human creativity, deception and ingenuity—qualities that attackers exploit with remarkable skill.
Why Existing Guardrails Are Not Enough
AI developers have invested heavily in safety layers, content filters and refusal mechanisms. Yet, no system is infallible. The interpretive nature of language means that malicious intent can be concealed within metaphors, fictional scenarios, coded instructions or indirect requests. As a result, even the most advanced AI models remain vulnerable to manipulation.
The limitations of current guardrails were highlighted in a 2025 European Union cybersecurity audit, which found that more than 40 per cent of tested AI systems could be tricked into generating restricted content through indirect prompting. These findings reinforce the need for a more comprehensive, multi‑layered approach to AI security—one that extends far beyond model‑level protections.
The Imperative for Layered Security
To curb the rise of malicious prompting, organisations must adopt a layered security strategy that integrates model‑level defences with infrastructure safeguards, governance frameworks and continuous monitoring.
At the model level, developers must prioritise training AI systems on adversarial datasets designed to anticipate harmful manipulations. Reinforcement learning processes must be continually updated to strengthen refusal behaviour, especially in the face of deceptive or ambiguous prompts. However, overreliance on internal safety mechanisms is dangerous. Recent incidents have shown that even highly sophisticated models can be misled by surprisingly simple techniques when attackers exploit subtle gaps in prompt interpretation.
Infrastructure‑level safeguards are therefore essential. Zero‑trust architectures, which assume that no input or user is inherently safe, can significantly reduce the risk of compromised outputs triggering harmful actions. Strict access controls, sandboxed environments and output‑verification layers ensure that even if an AI system produces an unintended response, it cannot directly affect critical systems or sensitive data.
Equally important is the separation of user inputs from system‑level instructions. Many successful prompt injection attacks exploit contexts where AI systems ingest external data—emails, documents, web pages or databases—without adequate sanitisation. By reinforcing boundaries between user‑provided content and system commands, developers can dramatically reduce the risk of indirect manipulation.
The Role of Monitoring and Behavioural Analytics
Continuous monitoring is a critical component of AI security. Advanced behavioural analytics can detect unusual patterns such as repeated probing, contradictory instruction sequences or escalating attempts to bypass restrictions. These systems act as early warning mechanisms, enabling organisations to intervene before a breach escalates.
A 2024 Gartner report predicted that by 2027, more than 70 per cent of large enterprises will deploy AI‑driven monitoring tools specifically designed to detect prompt‑based attacks. This shift reflects a growing recognition that AI safety is not a static achievement but an ongoing process requiring vigilance, adaptation and proactive defence.
Human Behaviour: The Often‑Overlooked Vulnerability
While technological safeguards are essential, the human dimension of AI safety cannot be overstated. Employees frequently expose AI systems to risk by entering sensitive data, ambiguous instructions or poorly structured prompts. A survey by Deloitte in 2025 found that nearly half of all AI‑related security incidents originated from user error rather than external attacks.
Organisations must therefore invest in comprehensive training programmes that educate staff about responsible AI interaction. Clear policies must be established regarding what information can be shared with AI systems, how prompts should be structured, and how to recognise potential manipulation attempts. Without this human‑centred approach, even the most advanced technical safeguards will fall short.
The Role of Policymakers and Regulators
Regulatory bodies have a pivotal role to play in shaping the global response to malicious prompting. Governments must develop comprehensive guidelines that define standards for AI deployment, data governance, model transparency and safety auditing.
Mandatory AI risk assessments, periodic audits and public safety reports should become standard practice for organisations deploying high‑impact AI systems.
The United Kingdom has already taken steps in this direction. The 2024 UK AI Safety Institute was established to evaluate emerging risks and develop global safety benchmarks. Similarly, the European Union’s AI Act introduces strict requirements for high‑risk AI systems, including transparency obligations and mandatory risk‑mitigation measures. These regulatory frameworks represent important progress, but global coordination remains essential.
Ethical Leadership and the Moral Imperative
Beyond technical and regulatory measures, there is a moral dimension to this challenge. Faith‑based leaders, community organisers and public intellectuals have a responsibility to advocate for ethical AI use. As societies increasingly rely on AI for communication, knowledge and spiritual engagement, ensuring the integrity of these systems becomes a matter of public trust and moral stewardship.
The misuse of AI—whether through malicious prompting, misinformation or manipulation—undermines human dignity and erodes the foundations of social cohesion. Leaders across all sectors must therefore champion technologies that reflect human values, protect vulnerable populations and promote the common good.
A Collective Responsibility for the Future
Curbing malicious AI prompting requires a society‑wide commitment. Developers, cybersecurity experts, policymakers, educators and end‑users must work together to build an ecosystem that prioritises integrity, resilience and accountability. No single organisation can address this challenge alone; it demands collaboration, transparency and shared responsibility.
The future of artificial intelligence will be defined not only by how advanced our models become but by how effectively we safeguard them. By adopting assertive, forward‑thinking strategies today, we can protect AI systems from manipulation, uphold ethical standards and harness the full promise of intelligent technologies for societal advancement.
The stakes are high, but the opportunity is greater. With decisive action, we can ensure that AI remains a force for progress, innovation and human flourishing in an increasingly complex world.
Professor Ojo Emmanuel Ademola, is the first African Professor of Cybersecurity and Information Technology Management, Global Education Advocate, Chartered Manager, UK Digital Journalist, Strategic Advisor & Prophetic Mobiliser for National Transformation, and General Evangelist of CAC Nigeria and Overseas