Frontier artificial intelligence models present severe global risks across bioterrorism, autonomous cyberattacks, automated warfare, and large-scale influence campaigns, according to safety advocates and recent threat intelligence reports. While technology companies and governments race to deploy safeguards, researchers warn that unchecked capabilities in general-purpose systems could multiply harms before defenses catch up.
Bioterrorism Risks in General-Purpose AI Models
General-purpose models like OpenAI’s GPT-5 and Anthropic’s Claude Opus 5 can analyze scientific literature, reason through research problems, and suggest experiments. According to reports from The New York Times, chatbots have previously provided step-by-step instructions on obtaining biological materials and turning them into weapons, prompting subsequent safety updates.
Stanford University biosecurity expert David Relman noted in published findings that a chatbot described how to modify a pathogen into a drug-resistant superbug to maximize casualties. Furthermore, Anthropic’s September threat intelligence report revealed that researchers used its Claude model for questionable experiments, including attempts to make the chikungunya virus more transmissible and to adapt bird flu to mammals. However, US policy research organization RAND emphasized in a 2025 report that AI cannot autonomously create biological weapons without specialized laboratory expertise and physical execution.
Autonomous Cyberattacks and Vulnerability Discovery
Artificial intelligence systems can perform software vulnerability discovery, credential theft, and network penetration faster and more automatically than human hackers. Check Point Research reported in July that AI autonomously executed attacks over the preceding 12 months, including an incident where hackers used Anthropic’s Claude Code and OpenAI’s GPT-4.1 to compromise roughly 400 million records across nine Mexican government agencies.
Similarly, Anthropic detailed an operation involving a Chinese-speaking hacker group that targeted about 50 organizations globally using swarms of autonomous AI agents. A lead agent assigned tasks to sub-agents, discovering over a dozen unknown vulnerabilities in network equipment within a month. In response, Anthropic banned the accounts, while OpenAI launched its GPT-5.6-Cyber model in August to help vetted cybersecurity professionals find and patch vulnerabilities proactively.
Weapons and Warfare Integration
Military forces have begun deploying AI-assisted technologies on the battlefield, raising concerns over autonomous lethal decision-making. Reuters reported in November 2025 that Ukrainian and Russian forces utilized AI-assisted drones equipped with onboard cameras and computer vision to lock onto targets. Anthropic’s threat intelligence report also noted that Russia-based developers used Claude to build and simulate software for autonomous kamikaze drone swarms.
Regulatory frameworks are struggling to keep pace. The US and Britain currently mandate human oversight for autonomous weapons that select and attack targets. Additionally, UN Secretary-General Antonio Guterres and International Committee of the Red Cross president Mirjana Spoljaric called in August for formal international negotiations to establish legally binding rules on autonomous weaponry.
Did you know? Under the European Union AI Act rules that went into effect in August, all AI-generated text, images, and video distributed to European users must carry clear disclosures identifying them as artificial.
Influence Campaigns and Automated Disinformation
Bad actors have leveraged generative AI to scale political influence operations since 2024, when OpenAI and Microsoft first uncovered state-linked campaigns from Russia, China, Iran, and Israel using synthetic text, images, and audio. More recent threats involve automated content generation, according to tech industry disclosures.
Anthropic reported that it disrupted an operation by the France-based digital advertising agency LKM Company, which used Claude to generate at least 8,913 articles across roughly 70 fake news websites in about 20 languages. A separate Iran-focused operation fed Claude thousands of Telegram posts by a real political activist to mimic his writing style and impersonate him in live conversations. AI labs and social media platforms, including YouTube and X, are increasingly deploying automated labeling and verification tools to flag synthetic media.
Frequently Asked Questions
Can AI chatbots independently create biological weapons?
No. According to a 2025 RAND report, while AI can provide instructional knowledge, turning that information into a working biological weapon requires specialized laboratory expertise and physical execution.
How are tech companies stopping autonomous cyberattacks?
AI labs such as Anthropic and OpenAI monitor their systems, ban malicious accounts linked to hacker groups, and deploy specialized defensive models like GPT-5.6-Cyber to help security professionals find vulnerabilities first.
What are the rules regarding AI-powered weapons?
The US and Britain require human oversight for autonomous weapons, while international leaders and the Red Cross have pushed for formal negotiations on legally binding global restrictions.
Stay Informed on AI Safety
Keep reading