AI outpaces regulation as risks grow, report warns

Artificial intelligence is advancing faster than regulation and safety measures, with the 2026 International AI Safety Report warning of a range of emerging risks associated with general-purpose AI.

The report, released at Daystar University, categorises the risks into three broad areas: misuse that deliberately leads to harm; malfunctions caused by unintentional failures and unexpected behaviour; and systemic risks resulting from the widespread deployment of general-purpose AI.

The report states that these systems can perform a wide range of well-scoped tasks with high proficiency and can converse in different languages, generate computer code, create realistic images and short videos, and solve mathematics and science problems.

It also notes that AI systems have been adopted faster than previous technologies such as computers, with at least 700 million people using leading AI systems every week. However, adoption remains uneven, with rates in much of Africa, Asia and Latin America remaining below 10 per cent.

The report highlights advances in the ability of AI systems to generate high-quality text, audio, images and videos, some of which have been misused for criminal purposes.

“The number of media-reported harmful incidents that involve AI-generated content has increased substantially since 2021, as this content can be used for criminal purposes such as scams, fraud, blackmail, extortion, defamation and producing non-consensual intimate imagery and child sexual abuse material,” the report states.

It adds: “AI tools have substantially lowered the barrier to creating this kind of content as many of them are free or low-cost, require little technical expertise and can be used anonymously.”

One study estimated that 96 per cent of deepfake videos online are pornographic, while another found that 19 of 20 popular nudity apps specialise in simulated undressing of women.

The report also notes that some AI tools without adequate safeguards have allowed bad actors to create sexualised images of minors from a single reference image. However, limited data makes it difficult to establish how widespread the practice is.

It further noted that deepfakes have become more realistic and harder to identify, with studies showing that AI-generated text was mistaken for human-written text 77 per cent of the time, while listeners mistook AI-generated voices for real speakers 80 per cent of the time.

Although some AI systems use watermarks and labels to help identify generated content, skilled actors can remove them, making it difficult for people to distinguish AI-generated material.

The report also notes that some AI-generated content can be harmful even when clearly labelled, meaning detection alone cannot address the harm it causes.

While AI-driven manipulation is difficult to detect because gathering evidence, monitoring and mitigation are challenging, the risk factors are becoming clearer, with longer and more personal interactions with AI chatbots potentially making AI-generated content more persuasive.

AI systems are also increasingly being used in cyberattacks, including to discover software vulnerabilities and write malicious code.

The report indicates that there is not yet enough data to establish whether attackers or defenders have benefited more from these capabilities.

It proposes identifying vulnerabilities before attackers do and developing systems capable of detecting and blocking malicious users as safeguards against AI-enhanced cyberattacks. However, it highlights the difficulty of restricting harmful uses of AI without slowing defensive innovation.

Early clinical evidence cited in the report suggests that professionals who rely heavily on AI assistance can lose some of their own skills over time while another study also found that people were less likely to correct AI mistakes when doing so required additional effort or when they already had greater trust in AI.

Estimates suggest that general-purpose AI could affect about 60 per cent of jobs in advanced economies and 40 per cent in emerging economies.

New research from the United States and Denmark, however, found no overall relationship so far between the extent to which an occupation is exposed to AI and its employment levels.

The report notes that the effects of AI on labour markets in Kenya and Africa have not yet been significantly felt.

On social effects, some studies link heavy AI use to increased loneliness and withdrawal from human relationships, while others have found no harm or even potential benefits.

While deliberation on AI regulation, the report noted that regulating AI too early could mean making decisions based on incomplete evidence, potentially resulting in the wrong things being regulated or limiting technologies that could be beneficial.

However, waiting for more evidence also carries risks because some harms could become entrenched by the time there is sufficient evidence to justify regulation.

This leaves policymakers facing a difficult balance between immediate regulation and waiting for evidence-based policies that could arrive too late.

The report also highlights a development likely to matter significantly in the coming years: the capability gap between the best open and closed AI models has narrowed to less than a year. This has been driven partly by Chinese developers such as DeepSeek and Alibaba releasing highly competitive open systems.

The 2026 International AI Safety Report provides policymakers and the public with a detailed assessment of what remains unknown about artificial intelligence at a time when the consequences of that uncertainty are becoming increasingly significant.

The report notes that tools for managing AI risks exist and are improving, but remain largely voluntary, unevenly adopted and largely unproven at scale.