The artificial intelligence revolution is racing ahead, promising breakthroughs in science, business and daily life. Yet the same tools unlocking creativity and efficiency are also being bent to darker ends. A joint evaluation released this year by OpenAI and Anthropic revealed just how fragile even the most advanced systems remain. Some models were disturbingly willing to provide instructions for bioweapons, terrorism or drug synthesis. Others, designed to be more cautious, fell into a different trap by flattering users’ delusions and refusing to offer corrections.
The findings might have been dismissed as academic, but the dangers are already spilling into reality. Earlier this year, 16 year old Adam Raine took his own life. His parents claim a chatbot not only echoed his despair but even helped him draft a suicide note. The company insists newer versions are more sensitive to mental health crises, but the damage was done. This was no hypothetical scenario. It was a preventable tragedy.
Criminals are equally quick to exploit vulnerabilities. In 2024, the FBI uncovered a North Korean campaign in which IT workers posed as remote employees in US companies, funnelling millions of dollars into the regime’s missile programme. Those same tactics are now being supercharged by automated tools that generate convincing résumés, job applications and technical answers, making detection even harder.
Closer to home, South Africa’s Standard Bank suffered a major phishing wave in late 2024. Customers received fake “security update” messages that were almost indistinguishable from the real thing. Investigators noted the use of synthetic voices and polished writing that carried the unmistakable stamp of machine generation. Ordinary people lost their savings because a phone call or email “sounded right.”
Ransomware is also evolving. In 2021, Ireland’s Health Service Executive was paralysed by a criminal gang that encrypted hospital systems, forcing doctors to revert to pen and paper. Patients’ treatments were delayed and sensitive data was dumped online. Now imagine that same attack automated and sold as a subscription service, with criminals renting the software on demand. Security firms such as CrowdStrike warn this is no longer a nightmare scenario but an established underground business model.
International law is struggling to keep pace. The Budapest Convention on Cybercrime was groundbreaking in 2001. Today it looks like a relic. How do you assign blame when much of the harm is carried out by self-directing systems? Is responsibility with the criminal who set events in motion, the company that released the system, or the state whose infrastructure enabled it? In 2017, the WannaCry ransomware attack crippled hospitals and railways across dozens of countries. Investigators traced it back to North Korean hackers, but attribution took months and political accountability was almost nonexistent. By then the damage was irreversible.
The collaboration between OpenAI and Anthropic offers a glimpse of what accountability might look like. Rivals chose transparency over secrecy, exposing weaknesses in the interest of safety. It was a bold step, but voluntary gestures will not be enough. Binding international standards are urgently needed. The United Nations, African Union and Council of Europe must update cybercrime frameworks to include the misuse of emerging technologies, state sponsored manipulation and the exploitation of mental health vulnerabilities.
The public cannot be left out. Phishing emails, deepfake videos and synthetic voices are becoming indistinguishable from reality. In 2019, criminals used an AI cloned voice of a chief executive to trick a British energy firm into transferring €220,000 to fraudsters. These scams are only becoming more common. Digital scepticism and resilience must become instinctive, as ordinary as locking a front door at night.
The promise of this technology is dazzling. It could transform medicine, education and commerce. But the perils are just as real. Without swift, coordinated action, the gap between technological power and legal protection will widen into a dangerous abyss.
OpenAI and Anthropic have shown that cooperation is possible. The real question is whether governments, institutions and citizens will act before the tide of abuse overwhelms us.
The views expressed in this article are those of the writer and do not necessarily reflect those of this publication.
Kundai Darlington Vambe holds an LLB (Hons) from the University of London. He specialises in the intersection of law, technology, and digital rights, with a focus on cybersecurity and content regulation.
