AI writer revisits his bet that open AI models won't wreck the internet

A writer at the newsletter Interconnects has revisited his claim that freely downloadable AI models would not wreck the internet. He still mostly stands by it, but he admits he is nervous. The timing is awkward. Last week was nicknamed "Black October" after a rush of critical security flaw disclosures, including ones called PHP2Shell and GitLab.

What he argued

Weeks ago, he wrote that an "open-weight" model (one whose inner workings anyone can download and run) would not cause a sudden, unprecedented disaster. He pointed to GLM-5.3, an open model he says matches the cyber ability once feared from Anthropic's Claude Mythos. Over a month after its release, he sees little public evidence of mass damage. At worst, he expects faster incidents, not a step change.

What the worriers said

He collected examples of language he found too strong. Anthropic's April 7, 2026 Project Glasswing post said the fallout for economies, public safety and national security "could be severe." Rob Wiblin said on April 10, 2026 that Mythos in criminal hands would be "an actual disaster." Zvi Mowshowitz, whom he called more reasonable, wrote that it "would be chaos." So far, he says, that chaos has stayed with companies running large digital products.

What the evidence shows

Epoch AI used GLM-5.3 to build a website that, when a Firefox user simply visited it, let an attacker pull files off that person's computer. The writer sees this as a lesson in how attacks work. Finding a flaw is one step. An attacker still has to plant a site that hunts for it, so several things must line up before real harm happens.

The Wall Street Journal reported that cyber first responders are strained. He reads that as the system working, with more people and money joining the effort to patch the internet. He also noted another major crypto security incident at Binance, though crypto has a long history of hacks.

What he now concedes

He is not a cyber expert and says he is making his best educated guess. He now thinks "fine" was the wrong word for what is coming. He describes a forced industry update to a new way of working with software, which will be expensive and hard. Many more critical flaws are being found in popular tools, and fixes are being built.

What to do

• Install updates on phones, computers and browsers soon after they appear.
• Keep business software, including tools like GitLab, current.
• Expect more frequent security notices over the coming months.