CrowdStrike and Fortinet co-sign a letter that dates the security tooling they sell

The person who has to deal with this is the security lead who gets the letter forwarded on Friday at 4pm with "thoughts?" attached. The document is usable for that, once you sort its recommendations by who has to move.

One pile needs nobody's permission, and it is the pile the letter puts first: a patch queue and an end-of-life inventory that were already losing the argument for change windows ^[5]. The other pile depends on someone else's goodwill. Governments are asked to expand trusted-access programs that hand organizations powerful models before commercial release ^[8], and the AI companies are asked to supply model access, funding, training and hands-on support, "especially for under-resourced critical-infrastructure defenders" ^[9]. A rural water district's frontier-model defence, in that design, arrives through an eligibility list it does not write.

Then count the signature page. Engadget's list names Cisco, Cloudflare and CrowdStrike ^[2]; TechCrunch adds Okta and Fortinet ^[6]. That is five incumbent security vendors ^[20] putting their names to the principle that "status quo security won't be enough" ^[11], and in most enterprises their products are the status quo the letter is describing. Read as candour, that is useful. Read as a renewal document, it hands the buyer one specific question, because the same letter tells cybersecurity companies to continuously test their defences against frontier model capabilities ^[7]: ask for those test results on the SKUs already on the invoice. Gizmodo reports that OpenAI, Anthropic, Google and Microsoft did not immediately respond to requests for comment ^[19], though the account team fielding your renewal call almost certainly will.

Teams often tell themselves the gap is one of detection sophistication, but the federal warnings describe something duller. In July the FBI and the EPA said attackers were targeting internet-connected programmable logic controllers at water and wastewater facilities, with utilities in at least seven states having reported incidents ^[18]. The later Siemens warning's novel element is the script generation; the reachability of the controller was already known ^[14]. The Five Eyes agencies made the timing argument without a product attached back in June, saying frontier models could transform offensive and defensive cyber capabilities within months ^[17].

So sort each ask into one of four boxes: does it need new spend, and does it need someone outside your company to act. No spend and self-contained is where a quarter of credible progress lives, and it is the box the letter's own first recommendation sits in. Spend and self-contained is the AI tooling pitch, worth a pilot scored on time to contain rather than alerts raised. No spend but externally gated is threat sharing and trusted access, worth applying to now because those queues only lengthen. Spend and externally gated is the vendor programs, including OpenAI's Daybreak, Anthropic's Mythos and Microsoft's Perception ^[12], and it can wait until the first box is honest.

The tradeoff, in the same breath: leading with the unglamorous box makes you look slow next to peers announcing AI pilots, and if the letter's "coming months" horizon ^[3] is right, a fully patched estate with no augmentation is still an estate defended at human speed. That is the bet I would take, on the grounds that the reverse bet leaves the reachable controller reachable.