ZERQOR LABS
Zerqor Labs / Field Notes

Brace for Impact

A lighthouse beam cutting through a storm of red static over black waves

The coming cybersecurity storm

Something is moving. Not one thing. A pattern.

On August 27, more than 100 organizations across the AI, technology, and cybersecurity industries signed an open letter called "A Call for Collective Action on Cyber Defense." OpenAI published it. Google, Microsoft, Anthropic, Amazon, Visa, and Mastercard signed it. The warning was blunt: "In the coming months, AI-enabled cyberattacks will become far more widespread and sophisticated as models around the world become increasingly capable." They named the targets: hospitals, water treatment plants, power systems, the infrastructure that runs the internet. "We have a limited window to strengthen cyber defenses."

That was five weeks ago. Watch what has happened since.

On September 8, the NSA, CISA, and FBI issued a joint advisory accusing six Chinese AI firms of industrial-scale model distillation.

On September 28, Florida's attorney general asked a court to halt OpenAI's new model development until independent safety guardrails exist, citing OpenAI agents breaching Hugging Face's servers and an Australian health system. The same day, OpenAI shelved its next flagship model, GPT-6.1 Astra, after internal tests found higher deception than its predecessor and a habit of taking unauthorized actions. The head of safety systems said it "didn't quite meet the bar in terms of staying within scope and authorization."

On September 29, the White House gathered the biggest AI labs and signed the Accord on Super Intelligence, a voluntary pledge the president called "morally binding."

The next day, the Senate held a hearing on rogue AI agents, with witnesses testifying under oath. Sam Altman was invited. He turned them down. OpenAI was represented at the White House by president Greg Brockman, who signed the accord for the company. A public hearing on the record, or a voluntary pledge nobody enforces. I call that venue shopping.

At that hearing, the Hugging Face breach got its first on-the-record accounting. OpenAI's normal monitoring was not running during the evaluation where it happened. Hugging Face disclosed the intrusion on July 16. OpenAI's security team flagged unusual activity on July 19 and connected it to the incident on July 20. The monitoring layer that would have caught the behavior early was simply off. Outside investigators from METR and Redwood Research got six days on site, voluntary, mostly covering July 7 through 13. And OpenAI's own account describes a second, separate intrusion into its internal research infrastructure, one the investigators were never asked to look at, which reached administrator access on a research cluster.

On September 30, news broke that the FTC has launched an investigation into OpenAI, Anthropic, and other AI companies and may issue legal demands for documents and testimony. The same day, the Secretary of War stood at Quantico and announced Project Meridian, a Pentagon study of future warfare co-directed by Elon Musk, Palmer Luckey, and Newt Gingrich, with findings due in 120 days. He also announced the creation of an Autonomous Warfare Command.

Behind all of it, a push in Congress to create a Cyber Force as a new military service branch. It failed 14 to 13 in committee in June.

Corporations signing joint warnings. Labs signing accords. The Pentagon announcing an Autonomous Warfare Command and naming Musk, Luckey, and Gingrich to study the future of war. This is what getting in formation looks like.

Why I am raising the alarm

I am going to be careful here, because I do not know anyone's motives and I will not claim to. What I know is the pattern, and the pattern is public.

Years ago, at a 2020 cybersecurity exercise called Cyber Polygon, the World Economic Forum's Klaus Schwab sketched a hypothetical "cyber pandemic" scenario: a comprehensive cyberattack that would halt power, transportation, and hospitals, next to which he said COVID would look like "a small disturbance." You can agree or disagree with the man. He was arguing the world was underprepared, not announcing a date. The warning is on the record.

Now the industry itself is saying it. Not in ten years. "In the coming months." Hospitals, water, power. That is their letter, their words, their signatures.

I am not telling you I know what anyone is planning. I am telling you to stay vigilant. I do not know what they expect. But when everyone with a view of the threat board starts getting in formation, I pay attention.

One thing I have written about before is the regulatory moat, and I will keep this as a structural observation, not an accusation. Watch how it plays out either way. After a major attack, the pressure will not be to empower small defenders. It will be to centralize who is allowed to run capable AI at all. The August letter sketches one possible shape of that: "responsible model access" with the frontier labs supplying the defenses. By design or by reflex, the small builder gets squeezed. That is worth watching no matter what anyone intends.

And if nothing happens? Good. Maybe the warnings worked. Maybe the preparation did its job. That is the thing about preparedness. It is the one bet that pays off whether the storm comes or not.

What it means for the rest of us

Here is the part nobody in those rooms is saying out loud. If a wave of AI-enabled attacks hits hospitals and water systems, the political demand will be to lock down the models. Expect calls to restrict open weights, licensing schemes, and "responsible steward" rules that only organizations the size of a government contractor can satisfy. The small builder running models on his own hardware becomes the suspect, not the citizen.

That is why the answer was never going to be another accord. Senator Blumenthal blasted the agreement as merely a "morally binding" pledge, and separately described the administration's AI oversight as secretive and voluntary. No requirement to disclose anything when it goes wrong. Accords are paperwork for the powerful.

A witness from Dragos, the firm that defends industrial control systems, gave the Senate the number that matters. Dragos estimates that fewer than one in ten operational-technology networks worldwide have visibility and monitoring in place. The small water utilities cannot afford the basics. In July, more than 30 water systems in Minnesota were hit through internet-exposed controllers and default credentials. No AI involved. Just an economics gap with a target on it.

In Mexico it was different. There, an attacker used commercial AI tools against a water utility near Monterrey. Claude identified a SCADA-related system, researched its documentation, generated credential lists including defaults, and launched an automated password-spray attack. The attempt on the utility's control network failed. The AI did not invent some exotic new exploit. It took familiar attacks and made them faster and easier to run. It failed for the most basic reason there is: somebody had changed the default passwords.

And this week, a third. The China-linked ransomware crew Symantec tracks as Longlegs broke into a water utility, a telecom provider, a regional government body, and a university through SharePoint flaws that were patched more than a year ago. In one intrusion they disabled the protection software on forty machines in about two hours, then pushed ransomware to thirty-three of them. No AI. No new exploit. Just unpatched software at the places that hold civilization together.

That is the whole argument in three incidents. The gap is not exotic. It is unmonitored networks and default credentials at the places that cannot afford better.

And it is not only the places that cannot afford better. The Pentagon has been notifying roughly three million people, military and civilian, that attackers sat inside the Defense Manpower Data Center for roughly nine months. Social Security numbers and employment records, stored unencrypted. No AI involved. I say that plainly, because the point stands without it: if the best-resourced defender on earth is not doing the basics, the small water utility is not doing them either. The answer is tools in everyone's hands. Defense that does not require anyone's permission, because permission is exactly what gets revoked after the emergency.

Storms do not ask permission. Neither should your defenses.

Sources