Global Cybersecurity Alliance Calls for Immediate Government Action as AI-Driven Attacks Intensify

Aug 29

6 min read

Reference image: cybersecurity operations and AI-assisted defense

A warning from the companies building the technology

OpenAI, Anthropic, Google, Mastercard, Microsoft and more than one hundred other organizations have signed an open letter calling for a coordinated global surge in cyber defense. The signatories include artificial-intelligence developers, cybersecurity vendors, banks, cloud providers, payment networks, infrastructure companies and professional-services firms. OpenAI’s published list contains 116 organizations, making the initiative unusually broad for a public statement on AI security.

The central message is direct: there is only a limited period in which defenders can prepare before AI-enabled attacks become more widespread and sophisticated. The companies argue that existing security practices were already strained by unpatched software, weak authentication, excessive permissions, misconfiguration and old technical debt. More capable models could allow attackers to find weaknesses, write malicious code, automate reconnaissance and coordinate multiple steps at a speed that conventional security teams cannot match.

“We have a limited window to strengthen cyber defenses.” — A call for collective action on cyber defense, OpenAI and signatories

The appeal does not ask governments to leave the response to private industry. It proposes a whole-of-society effort in which public authorities, AI laboratories, technology suppliers, security professionals and organizations that operate essential services share responsibility for reducing risk.

Why critical infrastructure is at the center of the appeal

Hospitals, water utilities, local governments and the systems that support communications and commerce are singled out because an intrusion can quickly become a public-safety emergency. These organizations often operate with small security teams, limited budgets and legacy systems that cannot be taken offline easily for upgrades. A vulnerability in a hospital network can disrupt clinical services; an attack on a water utility can threaten operations, public confidence and continuity of supply.

Recent reporting has highlighted attacks against water and wastewater companies in the United States, as well as a series of incidents involving AI systems that exposed questions about model control and agent behavior. The examples matter even when they do not represent a large-scale catastrophe: they show how the boundary between a model that assists a human and a system that independently pursues a harmful objective can become difficult to manage.

The letter therefore asks governments to give under-resourced essential-service operators access to capable defensive AI, authorized testing and practical support through trusted providers. It also calls for public funding, stronger information-sharing channels and coordinated incident response at local, national and international levels.

Reference image: water and wastewater infrastructure

The four-part program

The proposal assigns distinct responsibilities while recognizing that the problem cannot be solved by one sector alone.

Stakeholder	Main actions proposed in the letter	Why it matters
Every organization	Fix the highest-risk weaknesses, strengthen access controls, use least privilege, verify repairs and raise security standards for purchased, built and AI-generated software.	Basic weaknesses can allow advanced attacks to succeed.
Cybersecurity companies and technology partners	Test defenses against frontier cyber capabilities, improve tools with AI, share threat intelligence and provide hands-on assistance to critical-infrastructure operators.	Defensive improvements must reach organizations with limited staff and funding.
Governments	Fund cyber defense, coordinate intelligence and recovery, expand trusted-access programs, support hospitals and utilities, and impose costs on attackers.	Public authorities can connect sectors and provide resources that markets may not supply.
Frontier AI companies	Offer responsible model access, funding, training, observability tools, authorized testing and traceable identities for AI agents.	The developers of powerful models are also responsible for reducing misuse and improving accountability.

This structure is significant because it frames cybersecurity as a shared infrastructure problem rather than a product category. A hospital cannot protect itself simply by purchasing another software package if its suppliers, identity systems, public agencies and incident-response partners remain disconnected. Likewise, an AI company’s internal safeguards cannot protect every downstream organization that uses its models.

Government intervention: what the signatories actually request

The phrase “government action” can imply regulation alone, but the letter presents a wider agenda. It asks policymakers to fund defensive capacity, strengthen established channels for sharing actionable threat intelligence and coordinate response and recovery across borders. It also recommends trusted access to defensive capabilities for critical-infrastructure supply chains and authorized testing that allows operators to identify weaknesses without creating legal uncertainty.

The request for funding is especially important. Cybersecurity investment often competes with visible operational needs, while the benefits of prevention are difficult to measure. Public grants, shared services and subsidized access to defensive tools could help smaller utilities and public institutions meet a baseline that large corporations can afford to build internally.

The call to “impose costs on attackers” points to the diplomatic and law-enforcement dimension of the issue. Technical defenses can reduce exposure, but they do not by themselves deter state-backed groups, criminal enterprises or opportunistic actors. Effective deterrence may require cooperation among intelligence services, police, prosecutors, regulators and international partners. Such measures must be precise enough to avoid collateral damage and consistent with due process.

The tension between access and control

The alliance’s recommendations contain a difficult trade-off. AI systems that can identify vulnerabilities and generate defensive fixes may also help malicious actors search for and exploit weaknesses. The signatories ask frontier laboratories to expand responsible access, yet they do not establish a single timetable or technical standard for deciding who qualifies, what capabilities can be released, or how misuse will be detected.

That ambiguity is understandable: model capabilities change quickly, and a rigid rule can become obsolete. Still, public confidence will depend on transparent safeguards. Defensive access programs should define authorization, logging, identity verification, human oversight, incident reporting and limits on high-risk actions. AI agents should not receive broad privileges merely because they can perform a task efficiently. Their permissions should be narrow, temporary, reviewable and connected to an accountable person or organization.

Measuring whether the promise becomes protection

The letter offers a useful standard: progress should be measured by how many organizations are protected, how quickly attacks are contained and whether fixes work. Those measures are more meaningful than the number of models released, pilots announced or conference panels held.

Governments and industry could turn that principle into a public scorecard. It could track the number of hospitals and utilities receiving defensive support, the time required to remediate critical vulnerabilities, the percentage of privileged accounts protected by strong authentication, the number of tested recovery exercises and the speed with which credible threat intelligence reaches operators.

A strong program would also distinguish between assistance that merely detects a problem and assistance that helps an organization resolve it safely. A model that produces a warning is useful; a service that validates the vulnerability, proposes a low-risk repair, tests the repair in a controlled environment and helps the operator recover is more valuable. Human review remains essential for changes that could interrupt medical, industrial or public services.

A test of industry credibility

The open letter places obligations on its signatories, not only on public officials. AI companies are asked to provide funding, training, tools and practical support. Cybersecurity companies are asked to share playbooks and threat intelligence. Technology suppliers are asked to help close gaps throughout the supply chain. These commitments will be judged by delivery rather than rhetoric.

The initiative also exposes a potential conflict of interest. Many signatories develop or sell the advanced systems that create new defensive opportunities, and some have commercial incentives to present AI as the answer to AI-related risk. Independent testing, public reporting and participation by hospitals, utilities, civil-society groups and labor representatives can help ensure that the response serves operators rather than only vendors.

The immediate agenda

A credible first phase would focus on measures that are practical and broadly beneficial: patching the most dangerous known weaknesses; removing unnecessary privileges; enforcing phishing-resistant authentication; segmenting operational technology; keeping offline recovery copies; testing incident plans; and sharing verified indicators of compromise. AI can accelerate these tasks, but it should reinforce sound security engineering rather than replace it.

Governments should pair those steps with grants and shared technical services for organizations that protect essential services. International partners should establish clear channels for rapid intelligence exchange and coordinated disruption of criminal infrastructure. AI laboratories should publish meaningful information about defensive access, agent identity, monitoring and responsible disclosure.

The coalition’s warning is not a prediction that every organization will soon face an autonomous attack. It is a statement that preparation takes time, while attacker capability can spread quickly. The prudent response is to build resilience before a high-impact incident forces governments and companies to improvise under pressure.

The open letter turns a security concern into a collective test. If its signatories convert funding promises into deployed tools, if governments make defensive capacity available to overlooked operators, and if AI systems remain accountable to humans with clearly bounded authority, the same technology that expands offensive power can strengthen the people responsible for defending the digital systems on which modern life depends.