More than 100 artificial intelligence experts have published the second International AI safety report, warning that the overall trajectory of general-purpose AI systems remains ‘deeply uncertain’ despite rapid advances in capability and growing deployment.
Released earlier this month ahead of the India AI Impact Summit, the report outlines a broad spectrum of risks associated with AI systems. Some of these included potential impacts on jobs, environmental pressures and the malicious uses such as cyber-attacks and AI generated misinformation.
The authors note that while AI capabilities have improved over the past year, progress remains ‘jagged’, with systems excelling at some complex tasks while failing at simpler ones. They also highlight an ‘evaluation gap’, warning that benchmark results cannot reliably predict real-world utility or risk.
Sachin Agrawal, Managing Director for Zoho UK, commented: “AI is already delivering meaningful gains for UK businesses, from accelerating decision-making to strengthening fraud detection, and its potential will only grow as frontier systems mature. However, this very evolution, which the latest international safety report describes as ‘deeply uncertain’, shows why we need clear, enforceable regulation that gives confidence to innovators and the public.”
“AI requires strong transparency and governance practices, especially as cloud-based AI systems are increasingly developed in one country and deployed in another, making consistent oversight and responsible data handling even more important. With AI capabilities advancing far faster than traditional governance cycles, and many sectors relying on a small number of general-purpose models, measures such as clear documentation, monitoring and standardisation are becoming increasingly important.”
“Robust AI governance must go hand in hand with a firm commitment to data privacy and ethical management. As organisations adopt more advanced systems, they have a responsibility to ensure that the data powering these models is secured, transparent and protected. This approach won’t just help mitigate the risks noted in the report, such as unpredictable behaviour, deepfakes and misuse, it’s vital to earning long-term trust and delivering lasting economic and societal value.”
The report further highlights evidence gaps in understanding the prevalence and severity of AI-related harms, as well as uncertainty over whether current safeguards will be sufficient for more capable future systems.
It calls for more research into risk measurement, mitigation effectiveness and governance frameworks, noting that policymakers currently have limited visibility into how AI developers test and manage emerging risks.
The findings are expected to inform diplomatic discussions at the upcoming India AI Impact Summit, as governments consider how to balance innovation with safety, accountability and inclusive access to AI’s benefits.
