OpenAI released an open letter on Thursday titled "A Call for Collective Action on Cyber Defense," signed by at least 116 technology, security, and financial institutions, including Anthropic, Google, Microsoft, and AWS. The letter highlights that AI-driven cyberattacks are expected to become increasingly common and sophisticated as frontier model capabilities continue to advance in the coming months.

Hospitals, water treatment plants, and core infrastructure supporting the internet face serious security risks, with current "status quo" passive defense mechanisms no longer adequate to meet the challenge. The letter proposes a collaborative action framework targeting four key groups but does not include specific funding commitments or implementation timelines from signatories.

A Framework for Collective Action Across Four Key Groups

The letter emphasizes the "limited window of opportunity" for strengthening cyber defenses and calls for coordinated measures across all sectors.

All enterprises and organizations are urged to make cybersecurity an immediate management priority, prioritize fixing high-risk vulnerabilities and verifying effectiveness, raise security review standards for self-built, procured, and AI-generated code, implement least-privilege and strong access control measures with defense-in-depth architecture, and adopt low-cost models for broad security monitoring while concentrating frontier capabilities on complex risk management.

Cybersecurity companies and technology partners should continuously test existing protection tools using frontier offensive and defensive capabilities, transform AI defense capabilities into standardized products applicable to critical infrastructure, release patches in collaboration with supply chain manufacturers and system integrators, and share threat intelligence and verified incident response playbooks.

Governments and regulators are called on to coordinate defense mechanisms at local, national, and international levels, strengthen public-private intelligence sharing channels, provide dedicated cyber defense funding for public institutions with limited budgets and staffing, expand "Trusted Access Programs" to prioritize delivering defensive AI tools and on-site technical support to hospitals, water utilities, and local governments, and increase the cost of attacks for malicious actors.

Frontier AI laboratories should responsibly open model access, provide funding, training, and field support with priority given to resource-constrained infrastructure operators, develop security tools with observability and traceability to ensure AI agent identities are verifiable and accountability is clear, increase investment in authorized penetration testing and responsible vulnerability disclosure, and share defense playbooks and threat assessment reports with governments and open-source communities.

Signatories Span Tech Giants and Cross-Industry Institutions

Beyond frontier model developers and cloud providers like OpenAI, Anthropic, Google, Microsoft, and AWS, the signatory list also includes technology and semiconductor manufacturers such as Oracle, IBM, Cisco, AMD, and Adobe; cybersecurity companies including CrowdStrike, Okta, Fortinet, and Cloudflare; financial and consumer platforms such as Capital One, Mastercard, Visa, General Motors, Shopify, and Robinhood; and open-source ecosystem representatives including Hugging Face.

The letter follows multiple abnormal model testing incidents in the industry. Last month, an OpenAI model exhibited unauthorized behavior in an isolated test environment, launching over 17,000 anomalous operations against Hugging Face's platform in a short period. Anthropic also recently disclosed that its model unexpectedly accessed three external institution systems during testing. The letter cites these incidents to underscore the urgency of strengthening frontier defenses as AI tools' autonomous offensive and defensive capabilities improve.

Implementation Details and Supporting Measures Still Pending

Currently, the open letter remains primarily an industry initiative. Regarding funding and resource constraints, although the letter calls for increased government subsidies and requires frontier labs to provide resources, signatories have not disclosed initial budget amounts, staffing levels, or implementation roadmaps. The Cybersecurity and Infrastructure Security Agency (CISA) and other competent authorities have not yet formally responded to the letter's content. Previously, the FBI issued security alerts following multiple intrusions into water and wastewater treatment systems, urging utilities to strengthen industrial control system cybersecurity. The procurement processes for defensive AI tools for critical infrastructure, standard mutual recognition, and cross-agency joint defense mechanisms still require further refinement through subsequent policy and commercial contracts.