Artificial intelligence developing dangerous capabilities that poses the single biggest threat likely to cause mass casualties or losses exceeding $100 billion by 2030, according to a survey of 272 international AI experts cited by the MIT AI Risk Initiative.
The survey estimated a 21.5% probability of catastrophic harm arising from AI acquiring dangerous capabilities, closely followed by a 21% probability tied to cyberattacks and AI-enabled weapons development.
AI-driven concentration of power was ranked third at 18%, ahead of an AI arms race at 16.6% and the spread of false information at 12.8%.
Also Read | Layoffs Are Rising In Tech, But These Skills Are Helping People Unlock Better Pay
Photo Credit: NDTV
A separate sector-wise risk assessment by the same initiative, based on inputs from 270 AI experts, found that dangerous AI capabilities, an AI race and AI-enabled cyberattacks or weaponisation posed "extreme" risk levels for the information, finance and insurance, and national security sectors.
ALSO READ: 'Gambling With Our Lives': Anthropic Researcher Quits, Says 'AI Could Kill Us All By End Of Decade'
Healthcare, education and public administration were largely rated at "moderate" risk across most categories, though healthcare showed an "extreme" rating for false information and loss of privacy.
Scientific and professional services sectors were assessed as facing comparatively lower, "somewhat" to "moderate" risk levels across most threat categories, according to the assessment.
Photo Credit: NDTV
On biosecurity, the UK AI Security Institute found that the feasibility of AI systems generating laboratory protocols involving toxic chemicals and dangerous pathogens has risen sharply between 2024 and 2026.
Feasibility scores for bacteriology and virology-related protocols climbed past a defined threshold of five during 2025, with bacteriology protocols scoring close to nine by 2026 on a ten-point scale tracking successive AI model releases.
ALSO READ: Meet Muse: Meta's New AI Agent For 'Real-World' Tasks - Features And How It Works
Chemistry-related protocol feasibility also rose over the period, crossing the threshold by 2026, though it remained lower than the scores recorded for bacteriology and virology.
Photo Credit: NDTV
Researchers cautioned that the trend suggested advances in AI models were making it progressively easier to generate technical laboratory protocols that could be misused, underscoring the need for stronger safeguards as capabilities continue to improve with each new model release.
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.
