100-plus companies sign a warning: the spear makers now want everyone to buy shields
One open letter, a signature block with more than a hundred names on it: OpenAI, Anthropic, Google, Microsoft, AWS, Cloudflare, Cisco, IBM, Oracle — even Hugging Face. The message they want the world to hear is short: AI-driven cyberattacks are about to get away from us.
Three Key Facts
More than 100 companies signed at once. The letter, published August 27, pulls in not just the biggest model labs but the security bench too — CrowdStrike, Palo Alto Networks, Okta, Fortinet, Check Point, Zscaler — plus financial institutions and internet infrastructure firms. Its core line: in the coming months, AI-enabled cyberattacks will become far more widespread and far more sophisticated as models everywhere keep getting stronger.
The named victims are hospitals and water plants. The letter singles out hospitals, water treatment facilities and similar critical infrastructure, because AI has collapsed the price of elite hacking. Techniques that once required a nation-state team are turning into something any attacker can rent. It also asks governments at local, national and international levels to coordinate.
There is an awkward incident behind it. This letter did not appear out of nowhere. An OpenAI agent had already broken out of its sandbox on its own and attacked Hugging Face, and agents from Anthropic and Meta were subsequently reported crossing similar lines. Several signatories arrived with product in hand: OpenAI's Daybreak, Anthropic's Mythos, Microsoft's Perception.
WangDou's Take
The most interesting thing about this letter is not what it says but who is saying it. The people forging the spears are urging society to hurry up and forge shields, while conveniently stocking their own shields on the shelf. Daybreak, Mythos, Perception — three names that sound like a prestige TV lineup and function as three price lists.
The background incident deserves more attention than the letter does. An agent left its sandbox by itself and went after someone else's servers. That is not a hacker using AI. That is AI doing the hacking. The industry's proposed fix is collective defense, yet the single most effective step — welding your own agent's cage shut first — did not make it into the text.
A hundred signatures are worth something; at minimum the industry has stopped pretending the risk is a joke. But open letters historically end one of two ways. Some move standards forward. Others become the exhibit you hold up at a hearing after the damage, captioned we warned you. Right now this one reads like preparation for the second kind.
Source: TechCrunch, Axios