Big AI cashes in on its own risks

Welcome to The Prompt, your guide to the biggest issues and developments shaping AI, with Asia front and center. Want full access to all our AI reporting? Subscribe here.

Hello reader,

Big AI is cashing in on the cybersecurity threats it helped create in the first place.

Google has just announced Gemini 4 Carbon, demonstrating the model’s high benchmark scores in finding vulnerabilities, validating risks, and coming up with fixes. The company is initially limiting access to selected cyber defenders before expanding it through its Fairwind Program.

Like Anthropic’s Project Glasswing and OpenAI’s Daybreak, Fairwind gives approved organizations access to advanced capabilities to help secure their systems.

There’s nothing wrong with these efforts. As AI models become more powerful, defenders need better tools to keep up. Having a limited release means trusted defenders can get a head start while the labs refine safeguards, since the same capabilities could help attackers find and exploit security flaws.

But it’s very ironic that the companies pushing the boundaries of AI are also the ones selling protection against its risks.

Image credit: Made by Ulla/Tech in Asia with the help of AI

Oracle recently announced AI Security Mesh, which uses OpenAI’s cyber models to investigate weaknesses and help evaluate fixes. For OpenAI, that’s a way into enterprise security budgets through a vendor whose products companies already use.

The AI giants are targeting established security firms first, which already have the customers and distribution networks. By working with these firms, AI labs could profit from rising demand for cyber defenses without customer acquisition efforts eating into their revenues.

For now, the labs are absorbing some of the cost.

OpenAI has committed US$1 billion in subsidized Daybreak access over six months to help frontline defenders. Anthropic committed up to US$100 million in model usage credits when it announced Glasswing in April, and it has since expanded its verification program to let a wider range of security teams apply.

While the credits are temporary, the need for protection is urgent. The recent PHP episode shows how serious AI-led cyber attacks can get.

Analysts warned in May that Daybreak would likely sit alongside existing security tools rather than replace them, adding costs and integration work on top of what customers already pay.