Artificial intelligence (AI) is rapidly becoming powerful enough to significantly enhance the capabilities of criminals, terrorist groups and hostile states unless governments and technology companies strengthen safety measures, according to a new report published by the University of Cambridge's Centre for the Study of Existential Risk (CSER).
The report warns that increasingly capable AI systems could make sophisticated cyberattacks cheaper, faster and accessible to a much wider range of malicious actors while also accelerating online disinformation, fraud, biological and chemical threats, and the military use of autonomous weapons.
The warning comes as competition among leading AI developers has produced a new generation of so-called "frontier" AI models, prompting governments and intelligence agencies worldwide to reassess the technology's national security implications.
The Cambridge report comes at a time when companies including OpenAI, Anthropic, Google DeepMind, Meta and xAI are releasing increasingly advanced large language models capable of reasoning, writing software, conducting scientific research and completing complex tasks with minimal human supervision.
Unlike earlier AI systems designed primarily to generate text, today's frontier models can write and debug computer code, analyze vast datasets, browse the internet, use digital tools, plan multi-step workflows and collaborate with users on lengthy projects.
These advances are expected to transform industries ranging from health care and education to engineering and scientific research. AI is already helping researchers accelerate drug discovery, improve productivity and automate time-consuming tasks.
But the same capabilities that promise enormous economic and scientific benefits could also lower the barriers for malicious use.
According to the University of Cambridge report, AI is a classic "dual-use" technology—one capable of delivering profound benefits while simultaneously creating new opportunities for abuse.
The researchers stress that technological progress itself is not the threat; rather, the challenge is ensuring that safety measures evolve as quickly as AI capabilities.
The pace of progress has surprised even many experts.
Britain's AI Security Institute has found that AI performance in several areas, particularly cybersecurity, is improving at remarkable speed, with some cyber capabilities doubling roughly every eight months. Recent frontier models have already demonstrated expert-level performance on selected cybersecurity tasks.
Those advances have prompted growing concern among intelligence agencies.
Earlier this week, the Five Eyes intelligence alliance—comprising the United States, Britain, Canada, Australia and New Zealand—warned that AI models capable of enabling highly sophisticated cyberattacks against governments, businesses and critical infrastructure could emerge within months, calling cybersecurity an urgent leadership challenge rather than merely a technical issue.
-Cybercrime and misinformation top concerns
The Cambridge researchers identify cybercrime as perhaps the most immediate risk posed by advanced AI.
Powerful AI systems can automate the discovery of software vulnerabilities, generate convincing phishing emails in multiple languages, write malicious computer code, analyze stolen data and help coordinate increasingly complex cyber operations.
Although current AI developers have implemented safeguards to prevent many forms of misuse, researchers warn that increasingly capable systems could significantly reduce the technical expertise needed to carry out sophisticated cyberattacks.
Another major concern is the rapid spread of AI-generated misinformation.
As AI-generated images, videos and synthetic voices become increasingly realistic, experts fear that hostile governments, criminal organizations and influence operations could deploy fake content on an unprecedented scale to manipulate elections, deepen political divisions or spread false narratives before fact-checkers can respond.
The report also warns about the emergence of autonomous AI agents capable of carrying out extended sequences of actions with minimal human oversight.
Unlike conventional chatbots, these systems can independently interact with software, make decisions, execute tasks and coordinate multiple digital tools.
While such capabilities could greatly improve productivity, they may also reduce opportunities for humans to intervene if systems behave unexpectedly or are deliberately misused.
The researchers further caution that AI could increasingly assist research involving biological and chemical agents.
Although current models remain limited and have significant safeguards, future systems may offer scientific capabilities that require stronger oversight to prevent malicious exploitation while preserving legitimate medical and pharmaceutical research.
Rapid advances in AI-powered drones, surveillance systems and autonomous weapons are already reshaping modern warfare. Security experts warn that commercially available AI technologies could eventually be adapted not only by governments but also by non-state actors, increasing the speed, scale and precision of future conflicts.
Despite the growing risks, the report does not argue that AI development should be halted.
Instead, the University of Cambridge researchers recommend stronger cooperation between governments, AI developers and cybersecurity experts, greater transparency around frontier AI systems, wider adoption of rigorous safety testing before deployment and closer international coordination on AI governance.
The findings align with the broader International AI Safety Report 2026, which concludes that AI capabilities are advancing faster than many existing safety measures and identifies malicious use, technical failures and systemic risks as the technology's three most pressing challenges.
Ultimately, the Cambridge researchers argue that the world has a narrow window to prepare before AI capabilities become even more powerful.
"The same technology that can accelerate scientific discovery can also accelerate cybercrime," the report notes, emphasizing that preventing misuse will require continuous monitoring, international cooperation and security measures that evolve alongside increasingly capable AI systems rather than after serious incidents occur.
news_share_descriptionsubscription_contact
