OpenAI published an open letter on Aug. 28 calling for stronger global cyber defenses, signed by 116 organizations including Anthropic, AWS, Google, Microsoft, OpenAI and Oracle. The letter warns that as models improve, cyberattacks powered by artificial intelligence will become broader and more sophisticated, and that defenders are not ready.
CNBC, citing the letter, reported that the signatories argue “we have a limited window” to prepare. The letter describes attackers using AI tools to automate vulnerability discovery and phishing at a scale and speed that human defenders cannot match, while the defensive side still operates with tools designed for an earlier era of the internet.
The unusual breadth of the signatories is the story. OpenAI and Anthropic, competitors in the race to build frontier models, joined forces with the cloud companies that sell AI infrastructure and the enterprises that use it. The list reads like a who’s who of the American technology industry, and the shared message is that the threat is not hypothetical.
The letter marks a shift in how the AI industry talks about safety. For the past two years, much of the public debate centered on alignment: whether models could be steered away from harmful behavior. The new focus is operational: whether the infrastructure of the internet can withstand attacks that use AI against it.
Security researchers have documented the early contours of the threat. Phishing emails generated by language models are harder to detect than those written by humans, and automated tools can scan code for vulnerabilities faster than human auditors. Reports of AI-assisted intrusions have multiplied, and companies that sell defensive security products have begun marketing AI-powered defenses in response.
The letter’s authors argue that defenses must adopt the same technology as the attackers. That means AI systems trained to detect malicious activity, automated response tools that can move faster than a human security team, and shared information about attack methods across companies and governments. The signatories call for investment in cyber defense, better information sharing and international cooperation.
The timing is deliberate. Governments in the U.S., Europe and Asia are writing AI rules, and the letter is an attempt to steer that conversation toward security. Its backers want regulators to focus less on hypothetical future risks from superintelligent systems and more on the concrete problem of AI-enabled crime and espionage happening today.
There is also a commercial dimension. Companies that sell AI security products have an obvious interest in a more fearful market, and the letter’s emphasis on defensive investment aligns with the growth plans of Microsoft, AWS and Google, all of which offer AI security services. Critics of the letter may note that the signatories include companies that would profit from the very spending they recommend.
The letter is light on specifics, an omission its authors acknowledge. It does not propose a particular regulatory framework or a budget figure, and it stops short of calling for binding international agreements. Instead, it frames the problem in urgent terms and asks governments to treat cyber defense as a priority on par with physical security.
That framing has resonated within the industry. Executives from several signatory companies said in recent weeks that they are reorganizing security teams around AI, both to defend their own systems and to sell defenses to customers. The demand for security personnel with AI skills has outpaced supply, and universities are scrambling to add programs.
Concrete episodes have given the warning weight. Researchers have documented campaigns in which attackers used AI to draft convincing impersonation messages targeting corporate finance teams, and automated scanning tools have been observed probing cloud configurations for misconfigurations within minutes of new services going live. Security vendors say the cost of a single successful AI-assisted intrusion, measured in stolen data and ransoms paid, now routinely runs into the millions of dollars. The letter’s signatories argue that those numbers will only rise as models become cheaper to run and easier to obtain, and that the gap between what attackers can do and what defenders can stop is widening.
For the general public, the letter is a reminder of a quieter arms race. The AI models that generate impressive text and images are also being used to deceive, steal and disrupt, and the companies building those models are now publicly acknowledging it. The signatories’ credibility rests on the fact that they are also the primary targets: the data centers that train AI models are themselves among the most valuable attack surfaces on earth.
The letter’s authors said they intend to follow it with more detailed work, including technical papers and proposals for international cooperation. Whether that work produces action will depend on governments, which have moved slowly on cyber issues even as the threat has grown. For now, the message from 116 organizations is simple: the window is limited, and the defense has not caught up.
