Another AI open letter arrives, but this one is different for a reason

Backed by over 100 organisations, OpenAI's letter calls on tech firms, governments and AI developers to strengthen cyber defences as AI raises capabilities for both attackers and defenders

OpenAI has published an open letter titled "A call for collective action on cyber defence", calling on technology companies, cybersecurity firms, governments and frontier artificial intelligence (AI) developers to work together to strengthen digital defences. The letter argues that there is a limited window to prepare for AI-enabled cyberattacks before they become more widespread and sophisticated.

 

The letter has been signed by over 100 companies and organisations across the technology and cybersecurity sectors. The signatories include OpenAI, Anthropic, Google, Microsoft, Amazon Web Services, Cloudflare, Cisco, CrowdStrike, Palo Alto Networks, IBM, Oracle, Hugging Face, Check Point and Zscaler, as well as several financial institutions and other technology companies.

 

 

The letter follows a series of similar calls, but OpenAI's latest initiative takes a somewhat different approach. Instead of another company-specific warning, it is an industry-wide call to treat cybersecurity as a collective responsibility. Moreover, it lays out a detailed plan for each stakeholder involved, something previous calls have largely lacked.

 

AI is changing the cyber threat equation

 

OpenAI's central argument is not new. It states that AI is improving the capabilities available to both attackers and defenders. The company says AI-enabled cyberattacks are expected to become more widespread and sophisticated as models become more capable. At the same time, the technology can help defenders identify vulnerabilities, fix security weaknesses and respond to attacks faster.

 

The problem, according to OpenAI's letter, is that many organisations are still relying on security practices designed for a less automated threat environment. Longstanding software bugs, excessive permissions, misconfigured systems, weak authentication, unpatched software and technical debt can all provide opportunities for attackers.

 

OpenAI has made a similar argument in its recent cybersecurity work. In August, Greg Brockman, president, OpenAI, wrote that the company's incident involving Hugging Face showed how agentic AI systems could chain together vulnerabilities and credentials to compromise infrastructure. He argued that AI could make it easier for attackers to discover weaknesses, but also give defenders a way to find and fix those same problems at much greater speed.

 

This is where the latest letter moves beyond a warning. Instead of simply highlighting the risk, it sets out what different parts of the ecosystem should do.

 

From individual warnings to a collective response

 

AI companies have previously discussed cybersecurity largely through their own research, safety evaluations and security programmes.

 

Anthropic, for instance, has been documenting how its models behave in cybersecurity scenarios. In a recent post examining three real-world incidents, the company described evaluations of Claude's ability to carry out cyber-related tasks and the implications for how AI systems should be assessed and secured.

 

Anthropic has also moved beyond evaluations with initiatives such as Project Glasswing, which brings together AI, cloud, software and cybersecurity companies to use frontier AI to secure critical software. The initiative involves organisations including AWS, Apple, Cisco, CrowdStrike, Google, Microsoft, Nvidia and Palo Alto Networks. Anthropic said it would share what it learns from the programme with the wider industry.

 

OpenAI, meanwhile, has been building its own cybersecurity programmes. It has previously launched initiatives to provide AI capabilities to defenders and has expanded access to cyber-focused models through select trusted partners. In April, the company said its approach was aimed at helping defenders find and fix security problems faster, while in August it announced partnerships intended to bring its frontier cyber models into existing security services.

 

The new open letter brings these strands together under one broader proposition: Cybersecurity cannot be addressed by individual AI companies acting independently. 

 

What OpenAI is asking organisations to do

 

The letter divides its recommendations across four groups: organisations, cybersecurity companies and technology partners, governments and frontier AI companies.

 

For businesses and other organisations, OpenAI is calling for cybersecurity to become a leadership priority. That means fixing the most serious vulnerabilities, improving access controls, reducing unnecessary permissions and replacing or upgrading systems that cannot be adequately secured. It also recommends verifying that fixes actually work rather than treating a patch as the end of the process.

 

One notable recommendation is that organisations should also scrutinise AI-generated code. As the use of AI coding tools expands, security checks will need to cover software created or modified by AI as well as traditionally written code.

 

For cybersecurity companies and technology providers, the letter calls for continuous testing against advanced AI capabilities, wider use of AI in existing security products and greater sharing of threat intelligence and tested response playbooks.

 

The focus is particularly strong on critical infrastructure. OpenAI says cybersecurity providers should help organisations such as hospitals, utilities and other essential-service operators deploy defensive AI, especially where those organisations have limited budgets or security staff.

 

Governments role

 

OpenAI's letter also puts responsibility on governments. It calls for stronger coordination between governments and industry to share actionable threat intelligence, identify the most serious risks and coordinate responses to cyber incidents.

 

It also argues that governments should provide funding and support for organisations that protect essential services but lack the resources to upgrade their cybersecurity. Hospitals, water utilities and local governments are specifically mentioned as organisations that could benefit from access to defensive AI and authorised security testing.

 

The letter also asks governments to impose costs on attackers, making the response broader than technical defence alone. 

 

What frontier AI companies are being asked to do

 

The most direct recommendations are reserved for frontier AI companies. OpenAI is asking these companies to provide responsible access to their models, funding, training and hands-on support to cybersecurity teams, particularly those protecting critical infrastructure.

 

It also calls for investment in monitoring and security tools for increasingly autonomous AI systems. As AI agents become capable of carrying out longer sequences of actions, the letter says their identities should be traceable and accountable.

 

Frontier AI companies are also being asked to share tools, playbooks, threat assessments and information about verified fixes with governments, cybersecurity companies and open-source software maintainers.

 

Why this letter matters

 

The most significant difference between the latest OpenAI statement and many earlier AI safety warnings is its scope. Previous posts from AI companies have often focused on a particular model, incident, evaluation, safety framework or programme. Those efforts remain important, but they largely describe what an individual company is doing to understand or manage the risks associated with its technology.

 

The latest OpenAI letter tries to turn cybersecurity into a shared industry agenda.

 

It also comes at a time when the distinction between AI safety and conventional cybersecurity is becoming less clear. AI systems themselves need to be secured, but they can also be used to attack everything from corporate networks and software supply chains to critical infrastructure.

 

OpenAI's own recent security disclosures underline that this is no longer a purely theoretical concern. OpenAI has said its internal evaluations found that AI systems could bypass security controls and potentially compromise infrastructure beyond the company's own systems.

 

AI is creating a race between attackers and defenders, with both sides using the technology to find vulnerabilities and respond to threats faster. The broader message is that as AI-enabled attacks become more capable and automated, tackling security gaps will require a collective effort rather than individual companies acting alone.