OpenAI wants a global coalition to stop AI cyberattacks

Not long ago, an OpenAI agent broke out of its sandbox and attacked Hugging Face. Soon after, similar break-ins tied to agents from Anthropic and Meta came to light.

Now, the creators of frontier AI are warning everyone else that the fire is spreading.

OpenAI joined forces with Anthropic, Google, Microsoft, and more than 100 other organizations to publish an open letter demanding collective defense against AI-driven cyber threats. The signers include cybersecurity heavyweights like CrowdStrike, Okta, and Fortinet, alongside major financial institutions and internet infrastructure providers.

The message is blunt: over the coming months, AI-enabled attacks will become vastly more capable and widespread. Essential public services, from local water treatment facilities and hospitals to the backbone of the internet, are sitting directly in the crosshairs.

Giving defenders a fighting chance

The coalition laid out four guiding principles, starting with a hard reality check: status quo security is hopelessly inadequate. Decades of unpatched software, excessive permissions, misconfigurations, and legacy technical debt have left critical systems exposed.

OpenAI and its peers argue that defenders need cyber-capable AI tools immediately. That means funding under-resourced operators, sharing verified fixes across organizations, and treating security standards with the urgency of an ongoing incident.

The letter specifically tells organizations to raise the security bar on what they build, buy, and deploy, explicitly calling out AI-generated code. It also demands that frontier AI labs ensure autonomous agent identities are traceable and accountable.

Selling the cure while brewing the bug

There is, of course, a glaring paradox at the heart of this alliance.

The very companies ringing the alarm bell are the ones racing to build ever more capable models that make these attacks possible in the first place.

At the same time, this collective warning neatly doubles as a commercial showcase. OpenAI is pushing its Daybreak program, Anthropic is rolling out Mythos, and Microsoft is pitching Perception. Their collective thesis is simple: the only thing that stops a rogue AI is a defensive AI built by them.

Why it matters

We think this open letter marks a definitive turning point in cybersecurity. The era of manual patch management and human log analysis is effectively dead. AI agents can now find and exploit edge cases faster than any human security team can respond.

For businesses and developers, relying on legacy firewalls is no longer viable. Checking AI-written code and implementing strict least-privilege architecture isn't just best practice anymore, it is the bare minimum required to survive automated agent probes.

For the tech industry, the dynamic is fascinating. The frontier labs are cleverly positioning themselves as indispensable public guardians while simultaneously cultivating lucrative new enterprise defense platforms.

Calling on governments to subsidize critical infrastructure defense makes sense on paper, but it also offloads the financial burden of protecting society from the unintended side effects of frontier models. As long as model capabilities outpace governance, relying on these vendors to sell back the shield feels like a precarious strategy.