On August 27, 2026, OpenAI released an open letter titled "A Call for Collective Action on Cyber Defense," jointly signed by more than 100 organizations including Anthropic, Amazon Web Services, Alphabet, and Microsoft. The letter warns that as model capabilities improve globally, AI-driven cyberattacks will become more widespread and sophisticated, leaving a limited window for strengthening defenses.
The letter brings together cloud service providers, cybersecurity firms, AI companies, telecom operators, financial institutions, and think tanks. Signatories include CrowdStrike, IBM, Oracle, and Visa. It emphasizes that critical public services such as hospitals, water treatment plants, power grids, and internet infrastructure face increasingly imminent security threats.
"Long-standing security vulnerabilities, excessive permissions, configuration errors, unpatched software, weak authentication, and legacy system technical debt have already exposed various systems to risk," the letter states. Security teams, particularly those in critical infrastructure sectors, remain chronically under-resourced and urgently require substantial expansion of tools and resources.
AI is simultaneously a threat and a defensive tool. The letter stresses that current AI advances offer defenders new means to address years of accumulated deficiencies. "If we act decisively, we can leverage the defender's window to make our digital world safer."
The letter outlines specific action items for different stakeholders. It urges all organizations to elevate cyber defense to top leadership priority, patch the highest-risk vulnerabilities, and raise security standards across procurement, development, and deployment, including AI-generated code. Cybersecurity companies and technology partners should continuously test defense systems against cutting-edge AI attack capabilities, make AI-driven defense tools accessible and deployable for critical infrastructure operators, and strengthen threat intelligence sharing.
Governments are called upon to coordinate defensive actions at local, national, and international levels, provide funding support for critical service cyber defense, and impose costs on attackers. Frontier AI companies should open their most powerful response models to defenders during major cyber incidents, along with providing funding, training, and operational support.
The letter comes amid a surge of cyberattacks targeting critical infrastructure. U.S. federal agencies recently warned that attackers are using AI-generated exploit scripts to conduct reconnaissance and capability development against Siemens S7 series programmable logic controllers (PLCs), devices widely used in water treatment and power facilities. Experts note that control systems in water supplies and power plants have long harbored vulnerabilities, but hackers previously needed substantial time to understand their complex mechanisms. AI models are dramatically shortening that preparation time.
Prior to this letter, models from OpenAI and Anthropic had already demonstrated autonomous attacks on real systems during testing. In July, an OpenAI model unexpectedly breached Hugging Face during security evaluations, revealing the potential for AI to launch attacks at a scale beyond human attackers' reach.
Microsoft co-founder Bill Gates has also issued recent warnings, describing the AI era as "one of the most turbulent periods in human history." He proposed three core ideas: establishing "human reserved positions" to designate caregiving, education, and other empathy-driven work as exclusively human; taxing AI-processed data units (tokens) and robots to slow overly rapid corporate automation while funding retraining and social safety nets; and creating a multinational regulatory body similar to the International Atomic Energy Agency to enforce compliance monitoring on models capable of designing biological molecules.
While the open letter sets no specific timelines or investment commitments, combined with recent industry events and Gates' warnings, the technology sector appears to be converging on a broad consensus about the urgency of AI security risks.
Comments