What's this about?
Two days after the Hugging Face incident report and Trail of Bits' sandbox-escape writeup, more than 100 organisations put their names to a single public letter. The signatory list is unusually broad: the frontier labs (OpenAI, Anthropic, Google, Microsoft), the cloud providers (AWS, Oracle, Cloudflare), the security industry (CrowdStrike, Palo Alto Networks, Okta, Trail of Bits, Snyk, HackerOne), banks (Citi, Capital One, Mastercard, Visa, U.S. Bank), and Hugging Face itself. The core claim: AI-enabled attacks are about to become far more widespread, defenders currently have a window, and the window is closing.
What is a "defenders' window"?
Attack and defence both get better as models get better, but not at the same moment. Right now the strongest cyber-capable models are still held by a small number of labs, and open-weight equivalents have not arrived. That gap is the window: a period in which defenders can use AI to fix decades of accumulated bugs, misconfigurations and unpatched systems *before* the same capability is freely available to anyone who wants to attack. Once open weights catch up, the advantage flips. The letter's argument is that this window is measured in months, not years.
What's the community saying?
- The most concrete ask is the one drawing attention: governments should directly fund access to "capable defensive AI" for essential services like hospitals and water treatment, which do not have security budgets.
- Sceptics read the letter as the companies that built the offensive capability now asking the public to pay for the defence.
- Others note the letter is unusually specific for a joint statement — it names least privilege, defence in depth, and raising the security bar for AI-generated code, rather than staying at the level of principles.