OpenAI warns the 'defender's window' is closing as Chinese AI models catch up

OpenAI and a coalition have been patching critical PHP vulnerabilities found by AI since August, but the gap between U.S. frontier models and unrestricted Chinese open models is shrinking to about four months. The White House's new Super Intelligence Force coordinated the response, OpenAI's first major engagement with the body.

• Z.ai's Chinese GLM-5.3 trails U.S. frontier models by roughly four months
• OpenAI briefed CISA and the Super Intelligence Force on the PHP flaw
• Older mechanisms like the July Gold Eagle clearinghouse may already be outdated
• The next major internet bug may not allow a two-month defender's window