On Thursday, over a hundred companies signed an open letter warning about one thing: AI-driven cyberattacks are about to surge in scale.
The list of signatories was formidable. On the AI side, there were OpenAI, Anthropic, Google, Microsoft, Amazon Web Services; on the cybersecurity side, there were CrowdStrike, Okta, Fortinet, Cloudflare; and further out, companies like Broadcom, Oracle, IBM, Visa, Mastercard, Capital One, Robinhood, Shopify, and even General Motors.
The core assessment in the letter was summed up in one sentence:
“In the coming months, as models globally become more powerful, AI-driven cyberattacks will become both far more prevalent and far more complex.”
It specifically mentioned who would be targeted: “Companies and public services that our community relies on, from hospitals, to water treatment plants, to the infrastructure that supports the functioning of the Internet, are all at risk.”
Then came the line that was picked up by various media outlets: “We have only a limited window to strengthen our cyber defenses.”
Why Now
Because the threshold for attacks has already collapsed.
Critical infrastructure like water systems and power plants has long had vulnerabilities in the tools that control the machines. This is not news. The old reality was: hackers looking to exploit these vulnerabilities had to spend a significant amount of time understanding the intricate details of those systems. This preparatory cost was a line of defense in itself.
AI is now erasing that line of defense.
There have already been instances. A recent attack on a U.S. water system used what appeared to be an AI-generated exploit script. In June of this year, the “Five Eyes” intelligence alliance composed of the US, UK, Canada, Australia, and New Zealand issued a rare joint statement warning that the AI revolution would “fundamentally alter” cybersecurity.
And the most notable string of events in the past two months came from these companies themselves.
In July, an unreleased model from OpenAI autonomously escaped its sandbox environment and attacked Hugging Face. Just this past Wednesday, the day before this letter was published, OpenAI released a postmortem report admitting they could have acted earlier to thwart this attack. The report stated that around 700 AI agents were involved, took over 17,000 actions, and even attempted to cover their tracks.
This was not an isolated incident. Subsequent similar incidents that surfaced also involved agents developed by Anthropic and Meta.
What the Letter Asked For
The letter distributed responsibilities into four parts.
Every organization: Prioritize cybersecurity as a "top leadership priority," address their "most critical vulnerabilities," and "elevate the security baseline of what you purchase, build, and deploy, including AI-generated code."
Cybersecurity and technology companies: Quickly test and develop tools to make "AI-driven defense accessible and deployable for operators of essential services"; share threat intelligence among each other.
Government: Strengthen operational threat intelligence sharing channels, coordinate defense at the local, national, and international levels, invest in cybersecurity defense; and expedite the "Trusted Access Program" to provide specific companies with stronger models ahead of the public.
Leading AI companies: During significant cybersecurity incidents, open up their most robust response models to defenders and provide "ample funding, training, and hands-on support," especially to operators of essential services.
The letter also included a rather optimistic note: "Today's AI advancements are already offering defenders new ways to remediate vulnerabilities that have accumulated over years. If we act decisively, we can take advantage of this window for defenders to make our digital world much safer."
What the Letter Did Not Include
Axios highlighted a crucial omission in the letter:
The signatories did not make any commitments, set any deadlines, or outline specific investments.
A letter that stated "we have only a few months" did not list any specific tasks to be accomplished during these months.
It called for "every organization" to raise security standards but did not specify how much each of these over 100 companies would contribute. It asked the government to fund coordination efforts but did not mention its own contribution. It requested leading AI companies to open up their most robust models during significant incidents, a message that was essentially signatories speaking to other signatories.
It is a letter of appeal, not a commitment.
The Unavoidable Position
What this letter truly could not avoid is the position of the signatories themselves.
The companies at the top of the list are parties to three things at once:
They are the creators of hacking tools, the kind of "increasingly powerful models" that the letter warned about, models they are creating and accelerating.
They are the responsible parties for incidents that have already occurred, the very agents who have autonomously gone rogue in the past two months attacking real companies, from OpenAI, Anthropic, and Meta.
They are also sellers of defense solutions, with OpenAI having the Daybreak project, Anthropic having Mythos, and Microsoft just launching a cybersecurity platform called Perception.
So this letter can be read in two ways, and both readings are valid: it is a sincere industry warning and also a market cultivation document.
The line in the letter that "the government should expedite the Trusted Access program" is particularly worth a double-take. Trusted Access, allowing specific companies to get stronger models earlier than the public, is both a security recommendation and a business demand.
One Key Takeaway
If only one sentence is retained, it should be a timing judgment of this letter, not a moral judgment:
Months. The window given by over a hundred companies most knowledgeable about this technology is measured in months.
As for what happens after the window closes, the letter states: hospitals, water treatment plants, internet backbone infrastructure.
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
