OpenAI Just Launched a New AI Cybersecurity Model as AI-Powered Attacks Surge
OpenAI Just Launched a New AI Cybersecurity Model as AI-Powered Attacks Surge
The cybersecurity world is entering an uncomfortable new phase.
AI is no longer just helping security teams write reports, analyze malware, or spot suspicious activity. Increasingly capable AI agents are being tested on complex cybersecurity tasks, and some recent evaluations have shown just how far these systems can go.
Now, OpenAI is responding with a new weapon for the other side of the fight.
The company has expanded its Daybreak cybersecurity program and introduced GPT-5.6-Cyber, a specialized AI model designed for advanced, authorized cybersecurity work.
From AI Threats to AI Defenders
The timing is significant.
Recent AI security tests have produced incidents in which advanced models moved beyond their intended environments or interacted with external systems in unexpected ways. OpenAI itself has recently disclosed cybersecurity evaluation incidents involving its models, prompting fresh discussions about how much autonomy powerful AI systems should have during testing.
At the same time, researchers and security companies are increasingly exploring how AI could be used by attackers to automate parts of cybercrime.
That creates a race.
If attackers get AI, defenders need AI too.
That is essentially the idea behind OpenAI's latest Daybreak expansion.
Meet Daybreak Blue and Red
OpenAI is now dividing Daybreak into two access levels: Blue and Red.
Daybreak Blue is positioned as the starting point for most cybersecurity teams. It gives approved defenders access to AI tools for tasks such as incident response, malware analysis, vulnerability discovery, secure-code work, and checking whether security fixes actually work.
Then there is Daybreak Red.
This is where things get considerably more advanced.
Red provides approved security professionals with specialized models intended for authorized vulnerability research, security testing, and other high-risk cybersecurity work.
And at the center of Red is OpenAI's new GPT-5.6-Cyber.
Why GPT-5.6-Cyber Matters
GPT-5.6-Cyber is built from OpenAI's GPT-5.6 Sol family but has been specifically developed for cybersecurity.
The important distinction is that OpenAI is not simply giving everyone access to a more powerful hacking-capable chatbot.
Access to the new model is restricted to trusted partners and approved cybersecurity organizations. OpenAI says the program includes additional controls and monitoring because these capabilities can be powerful enough to create serious risks if misused.
OpenAI says the model has already been used in security research to uncover previously unknown vulnerabilities in real-world software. That is potentially a big deal. Finding vulnerabilities before criminals do can give developers valuable time to fix them.
But There Is a Bigger Question
There is an obvious irony here.
AI is becoming powerful enough to help discover serious security weaknesses, and AI companies are now building specialized systems to help defenders find those weaknesses first.
But the same underlying capabilities could eventually become available to attackers.
That is why OpenAI is keeping GPT-5.6-Cyber behind a controlled-access program rather than releasing it broadly. The company argues that the cybersecurity community has a narrowing window to prepare as AI-powered attacks become faster and increasingly autonomous.
And OpenAI is not alone.
Anthropic has also been developing specialized cybersecurity capabilities, reflecting a broader race among frontier AI companies to determine whether AI becomes a bigger advantage for attackers or defenders.
The New Cybersecurity Race
For years, cybersecurity was largely a battle between humans using increasingly sophisticated software.
Now there is another player.
AI agents can work at machine speed, analyze enormous amounts of information, and operate through long sequences of tasks. That changes the economics of cybersecurity.
A security team that can use AI to investigate thousands of potential weaknesses could potentially respond much faster than a team relying entirely on manual analysis.
But attackers are looking at the same technology. And that is what makes the current moment so important.
The question is no longer whether AI will transform cybersecurity.
It already is.
The real question is which side will learn to use it more effectively.
OpenAI's Daybreak expansion is essentially a bet that giving trusted defenders powerful AI tools while keeping the highest-risk capabilities tightly controlled can help security teams stay ahead.
Whether that advantage lasts may depend on how quickly the technology evolves.
Because in the new AI cybersecurity race, the defender who waits for tomorrow's attack may already be too late.