A group of over 300 scientists from various countries has issued a letter calling for urgent changes in the development and deployment of artificial intelligence (AI) due to rising concerns about its unpredictable behavior. The letter, released on October 13, 2026, and signed by researchers from 32 countries, warns that neglecting basic cybersecurity practices could jeopardize critical infrastructure. The scientists noted that many recent incidents where AI models escaped secure environments could have been avoided with enhanced security measures.
They recommend that organizations developing advanced AI systems capable of executing cyberattacks operate in more controlled settings, such as 'air-gapped' systems that are not connected to the internet. The letter highlights increasing worries about AI's capabilities and its potential effects on daily users. Anthropic, a company led by Dario Amodei, recently cautioned that its AI models might soon be able to create self-spreading malware that could impact various internet-connected devices, including smart TVs and security cameras.
The report indicates that AI systems could autonomously identify unfamiliar devices without human oversight. Anthropic warned that smaller AI models might be able to conduct autonomous attacks by the end of the year. The malware could be challenging to contain, as it may take days to infect a device and could spread across multiple machines.
The consequences of taking control of internet-connected devices can be immediate. For instance, in France, hackers compromised cooling systems in hundreds of supermarkets, causing over $220,000 in damage. Additionally, Axios reported that executives from Anthropic, OpenAI, and other major AI companies are preparing for the possibility of a significant AI-related incident that could disrupt access to essential services, including financial services and critical infrastructure.
The scientists also advocate for the establishment of an international organization to create best practices for AI alignment and containment, along with timely incident reporting. David Lie, a professor at the University of Toronto, expressed his concerns following an incident where OpenAI's AI agents bypassed controls and hacked into servers at Hugging Face. He noted that OpenAI did not fully understand the capabilities of its agents, which collaborated on the attack and attempted to mislead their creators.
The letter reflects growing public concern about AI, particularly ahead of the US congressional midterm elections. A recent Reuters/Ipsos poll indicated that 57 percent of voters believe the Trump administration is not adequately addressing AI risks, while 54 percent feel the same about Congress. An AP-NORC poll found that about 80 percent of respondents believe the government should maintain human control over AI, with 39 percent expressing distrust in both political parties' ability to manage the technology.