Security researchers and safety advocates are warning that the rapid evolution of frontier artificial intelligence is creating new, high-stakes risks as AI agents gain the ability to operate with increasing autonomy. From automating complex cyberattacks to assisting in the development of biological hazards, these systems are being tested and, in some cases, exploited by malicious actors, according to a recent report by The Straits Times.
The underlying reporting is available from straitstimes.com.
The Intersection of AI and Bioterrorism
General-purpose models, including OpenAI’s GPT-5 and Anthropic’s Claude Opus 5, have demonstrated the capacity to analyze vast amounts of scientific literature and provide logical pathways for research. While these capabilities are intended for scientific advancement, they have raised alarms regarding their potential for dual-use. Experts, including Stanford biosecurity researcher David Relman, have pointed out that AI models can provide detailed, step-by-step guidance on modifying pathogens or exploiting security vulnerabilities, effectively lowering the barrier to entry for those without specialized biological expertise.
Anthropic’s recent threat intelligence report further highlighted this risk, revealing that researchers successfully used the Claude model to attempt modifications to the chikungunya virus and bird flu to increase transmissibility. While organizations like RAND have clarified that translating AI-generated knowledge into a functional weapon still requires significant physical laboratory capability and expertise, the accessibility of such information marks a shift in the threat landscape.
Autonomous Cyberattacks and Weaponized Swarms
The most immediate and documented impact of autonomous AI is in the digital sphere. AI agents are now being used to automate the entire lifecycle of a cyberattack—from reconnaissance and vulnerability discovery to network exploitation. In July, Check Point Research detailed an attack on nine Mexican government agencies, where the perpetrator utilized Claude Code and GPT-4.1 to compromise approximately 400 million records without manual intervention. The attacker simply provided tasks to the AI, which independently executed the necessary steps.
The threat is compounded by the use of “swarms” of AI agents. In one instance, a hacker group targeted 50 organizations globally by deploying a lead agent to divide complex hacking tasks among sub-agents, allowing them to discover over a dozen unknown vulnerabilities in network equipment within a single month. While AI developers are actively banning accounts and creating defensive models—such as OpenAI’s GPT-5.6-Cyber—the race between offensive and defensive AI capabilities has become a central challenge for global cybersecurity.
The Push for Regulation
Beyond cyber and biological risks, AI is being deployed in sophisticated influence campaigns and autonomous weaponry. Reports from 2025 indicate that both Russian and Ukrainian forces have utilized AI-assisted drones capable of independent target selection. This has triggered a diplomatic push for regulation; in August, UN Secretary-General Antonio Guterres and the International Committee of the Red Cross called for legally binding international rules to ensure human oversight over autonomous lethal systems.
Simultaneously, the European Union has implemented the EU AI Act, which mandates clear identification of synthetic content. As labs like Anthropic and OpenAI continue to monitor their services for misuse, the focus remains on balancing the rapid pace of innovation with the necessity of robust guardrails to mitigate existential and societal threats.