Nathan Lambert Says GLM-5.3 Release Produced No Attack Wave

Nathan Lambert argues the open-weight cyber-risk conversation has broken down around a prediction that keeps failing to land. In a post at Interconnects dated October 6, 2026, he writes that "By all measures, GLM-5.3 is the model that crosses that threshold of capability, and there's little public evidence that much has changed, and we're over a month out of the model weight release."

The frame he is pushing back on comes from Anthropic, which has been a near-daily presence in our tracker over the past 90 days. When the lab previewed Claude Mythos, Lambert writes, "it was previewed as if it was a new class of cyber weapon, where if it ended up in the wrong hands it would've caused mass societal destabilization. It seems like this was wrong." He goes further: "If Claude Mythos was accidentally released as open-weight, it seems like the world would have been more or less fine."

Lambert's core policy argument is one of symmetry. "If you think the latest open-weight models need to be banned to slow the diffusion of cyber risks, you probably also need to make public-facing APIs for the frontier closed-models illegal," he writes, placing himself with Hugging Face and Joshua Saxe in what he calls the AI risk moderates camp. He also pushes back on the characterization that Chinese labs are reckless, noting that "Running comprehensive safety evaluations on a frontier model like Kimi K3 could cost tens of millions of dollars in compute. These labs would much rather leave that for training."

He names what he is doing. The warning that open weights will cause substantial harm is, in his words, "a falsifiable prediction," and so far it has not been borne out.