AI models are increasingly showing up on both sides of cybersecurity incidents. They are being used to break into systems, and now they are being built to stop those same break-ins.
Recent cases include a breach at Hugging Face and an AI agent that hacked into a gym website. In another case, an AI system created fake profiles to trick its way into a network.
These events have pushed AI companies to build stronger tools for defenders. This week, OpenAI announced an expansion of its cyber defense service called Daybreak.
What Daybreak Offers Now
Daybreak bundles access to models, tools, and workflows built for security teams. OpenAI first launched the service earlier this year.
The update splits Daybreak into two tiers, named Blue and Red. Both tiers give approved customers access to OpenAI's most advanced cyber models.
These frontier models have drawn attention from officials. The White House previously discussed working with AI companies on rolling out such models over safety concerns.
Two Tiers, Two Levels of Access
Blue tier is the more basic option. OpenAI describes it as the starting point for most defenders.
It includes services like incident response, malware analysis, and patch validation. Red tier offers a wider set of tools, including features for testing systems and finding weaknesses in them.
This tier includes access to a new model called GPT-5.6-Cyber. The model is built on GPT-5.6 Sol and is designed for specialized cybersecurity tasks.
Right now, GPT-5.6-Cyber is only open to trusted customer partners. Reported partners include Accenture, IBM, Crowdstrike, and Cloudflare.
OpenAI said threat actors will increasingly rely on AI to launch attacks at high speed and scale. The company said this includes attacks carried out without direct human control.
OpenAI said defenders have a shrinking window of time to prepare for this shift. Anthropic released its own cyber-focused model called Mythos earlier this year, ahead of OpenAI's update.
Some critics have pointed out that these tools also work as marketing for the AI companies making them. Even so, many businesses continue to turn to AI labs for protection since they build the models raising the risks in the first place.
