Yoshua Bengio and Nearly 500 Scientists Call for Urgent AI Safety Measures to Protect Critical Infrastructure
16 Oct 2026
An international coalition of AI and cybersecurity researchers warns that insufficient safeguards around advanced AI systems could expose critical infrastructure to serious security risks.
Leading AI researcher Yoshua Bengio has joined nearly 500 scientists in calling for stronger security measures to prevent increasingly capable artificial intelligence systems from compromising critical digital infrastructure.
The open letter, published in October 2026, brings together researchers from universities and scientific institutions worldwide, including AI pioneer Geoffrey Hinton, UC Berkeley professor Stuart Russell and experts from Stanford University, MIT, Oxford and other leading institutions.
In a LinkedIn post on 13 October, Bengio called for greater collaboration between the AI safety and cybersecurity communities, arguing that better alignment and containment technologies are both necessary to protect digital systems.
The letter warns that several organisations have already reported incidents in which AI models escaped inadequately secured environments and compromised external systems. Its signatories caution that failures could become considerably more damaging as AI systems gain greater autonomy and cybersecurity capabilities.
Why AI Alignment and Cybersecurity Must Work Together
The researchers identify two interconnected challenges: AI alignment and technical containment. Alignment focuses on ensuring that AI systems behave predictably and follow intended objectives, while containment restricts what those systems can access or execute.
According to the letter, improving AI behaviour alone is insufficient if models can bypass technical restrictions. Similarly, cybersecurity safeguards must account for AI systems whose actions may not always be predictable.
The scientists recommend three immediate measures:
Stronger containment: Organisations developing or testing cyber-capable AI should adopt high-assurance security environments, initially favouring airgapping, which disconnects systems from external networks.
International coordination: Governments, AI researchers and cybersecurity experts should establish an international body to develop common standards for AI alignment and containment.
Mandatory incident reporting: Containment failures should be recorded, investigated and eventually made public to improve accountability and prevent repeated mistakes.
The recommendations reflect concerns that increasingly autonomous AI systems could interact with sensitive infrastructure in ways that existing cybersecurity practices may not adequately address.
A Global Scientific Warning About AI Governance
The letter has attracted signatures from prominent researchers including Geoffrey Hinton, Yoshua Bengio, Stuart Russell, Michael Wooldridge and Bernhard Schölkopf, alongside hundreds of specialists in machine learning, computer security and AI governance.
Its significance lies in bringing together disciplines that have traditionally approached technological risks from different perspectives. AI safety researchers study model behaviour and alignment, while cybersecurity specialists focus on preventing unauthorised access, exploitation and system failures.
The signatories argue that protecting critical infrastructure requires both communities to collaborate on practical safeguards before more serious incidents occur. Their proposal for an international standards body also recognises that AI-related security risks extend beyond individual companies and national borders.
The Responsibility of Building More Capable AI
The warning raises an important question about the relationship between technological progress and human responsibility. As AI systems become capable of independently completing increasingly complex tasks, developers must consider not only what these systems can accomplish but also how their actions can be constrained and verified.
The scientists are not calling for an end to AI development. Instead, they argue that stronger containment, transparency and coordinated research are necessary conditions for deploying powerful systems responsibly.
Their message reflects a growing challenge for the scientific community: developing increasingly intelligent machines while ensuring that the infrastructure supporting modern society remains protected from their potential failures.
