Greg Brockman, a co-founder of OpenAI, issued a blunt warning this week: the window for securing AI systems is closing rapidly. He described an escalating arms race between defenders and attackers, and said that failing to deploy security tools quickly enough will create serious risks down the road.
An Arms Race Picking Up Speed
Brockman, who helped build the company behind ChatGPT, painted a picture of a contest that's accelerating on both sides. Every new safeguard seems to be met with a fresh workaround. Attackers are studying AI models as fast as researchers can patch them. The old rhythm of software security — test, release, patch, repeat — doesn't hold up when the technology keeps changing shape.
He didn't name specific incidents or companies. He didn't need to. The point was about the pace itself. Defenders are often one step behind, and that gap is getting harder to close.
Why Waiting Is a Bad Bet
Brockman's core argument is that waiting until a vulnerability is exploited before acting is no longer a viable strategy. Security tools need to be deployed early, even if they're imperfect. The alternative, he warned, is a future where AI systems are compromised before they're fully understood or trusted.
He stressed the timing aspect — not just building defenses, but getting them into the field. A tool that sits in a lab for months while the threat landscape shifts is almost as risky as having no tool at all.
A Warning From Inside the Industry
Brockman's voice carries weight because he's not an outside critic. He's a leader at one of the most prominent AI labs in the world. When someone with that vantage point says the window is closing, it's worth taking seriously.
He didn't offer a specific deadline or a detailed roadmap. Instead, he framed it as a collective problem — one that the entire field needs to confront before it's too late.
The question now is whether the industry will move faster. Brockman has made his position clear: the time to act is now, not after the next breach.
