Was I wrong about open models & cyber risks?

| Source: Interconnects (Nathan Lambert)

Tags: Nathan Lambert, GLM-5.3, Claude Mythos, Open-Weight Models, Cybersecurity, Interconnects

Nathan Lambert argues that open-weight GLM-5.3 already crossed the cyber-capability threshold once feared for Claude Mythos, with little public sign of a step change in harm, suggesting open models accelerate rather than transform cyber risk.

Details

Writing after a wave of critical vulnerability disclosures he calls Black October, including PHP2Shell and GitLab, and a major Binance security incident, Nathan Lambert revisits his view on open models and cyber risk. He argues GLM-5.3 is the model that crossed the feared capability threshold, yet over a month after its weights were released there is little public evidence of dramatic change. He goes further: had Claude Mythos leaked as open-weight, the world would likely have seen more incidents but an acceleration rather than a step change. He points to reports of strained cyber first responders as a sign the defensive system is working, while expecting vulnerabilities to keep rising for some time.