The AI race for cybersecurity just got more intense with GPT-5.5-Cyber

The battle between AI giants is expanding beyond chatbots and productivity tools. Now, cybersecurity has become one of the industry’s fastest-moving battlegrounds, with OpenAI unveiling GPT-5.5-Cyber as its latest push into defensive security work.

OpenAI said GPT-5.5-Cyber is rolling out in a limited preview to defenders responsible for securing critical infrastructure. The company said the model is intended to support specialized defensive workflows while continuing to block requests that could enable real-world harm.

The move comes just weeks after rival company Anthropic expanded access to its own cybersecurity-focused offerings. Anthropic recently launched a public beta of Claude Security for enterprise customers and previously previewed a heavily guarded model called Mythos.

OpenAI is loosening restrictions for security experts

Unlike the standard GPT-5.5 model, which has broader safety limitations, GPT-5.5-Cyber is intentionally more permissive for approved users. OpenAI said the goal is to make cybersecurity workflows more practical for trusted defenders who need advanced testing capabilities.

The company explained that the rollout will happen gradually through tiered access levels. OpenAI also stressed that the model is not necessarily stronger than GPT-5.5 in every cybersecurity benchmark, but instead focuses on enabling specialized defensive tasks with tighter monitoring and identity verification.

According to OpenAI, both individuals and businesses can apply for access by verifying their identity or working through enterprise representatives. The company also highlighted partnerships and feedback from firms including Intel, SentinelOne, Okta, and Snyk.

Anthropic’s security push is already showing results

Anthropic has been building out its cybersecurity offerings in recent months. Its Claude Security product is designed to help enterprise customers scan code for vulnerabilities, validate findings, and generate targeted patches for teams to review in Claude Code.

Anthropic has said Opus 4.7 is the first broadly released model on which it is testing its newer cyber safeguards, while Claude Mythos Preview remains a tightly restricted research model under Project Glasswing. Anthropic has described Mythos Preview as unusually capable at finding and exploiting software vulnerabilities during internal testing.

One of the biggest examples came from Mozilla, the nonprofit behind Mozilla Firefox. Mozilla said its early access testing of Claude Mythos helped uncover and fix 271 vulnerabilities during an initial evaluation period.

Mozilla chief technology officer Bobby Holley said the volume of flaws identified by Claude Mythos Preview created a sense of “vertigo” for the Firefox team. Mozilla later shared more of its AI security work by open-sourcing its 0DIN AI Security Scanner and publishing details about how it evaluates AI-related security issues.

Little-known fact: A 2025 industry survey found that more than 51% of organizations already use AI tools in at least some cybersecurity operations.

AI is changing how software vulnerabilities are discovered

The rise of models like GPT-5.5-Cyber and Claude Security highlights how quickly AI is transforming cybersecurity work. Traditionally, vulnerability research and defensive testing required large teams of experienced engineers working through software manually.

AI systems can now scan large codebases, identify potential weaknesses, validate findings, and generate targeted fixes more quickly than traditional manual review alone. That could help companies shorten the gap between discovery and remediation as software ecosystems become more complex.

At the same time, these capabilities also create new concerns. The same systems that help defenders identify vulnerabilities could potentially be misused if safeguards fail or restricted tools fall into the wrong hands. That is why companies like OpenAI and Anthropic are limiting access through verification systems and enterprise controls.

The cybersecurity AI race is only getting started

The growing competition between OpenAI and Anthropic shows that cybersecurity is becoming one of the most important use cases for advanced AI systems. Instead of focusing only on chat assistants or creative tools, AI companies are increasingly targeting enterprise security operations, where demand is rapidly increasing.

As more organizations experiment with AI-powered vulnerability discovery, defensive automation, and code analysis, the pressure will likely intensify for companies to deliver stronger security-focused models. The early results already suggest these tools could reshape how software vulnerabilities are found and fixed across the industry.

Little-known fact: Cybercrime damages worldwide are projected to hit $13.82 trillion annually by 2028, up sharply from previous years.

A new era for defensive AI tools

The launch of GPT-5.5-Cyber signals that the AI industry’s cybersecurity arms race is accelerating quickly. OpenAI and Anthropic are both trying to prove that their models can become essential tools for defenders without opening the door to dangerous misuse.

For security teams, the technology could dramatically reduce the time needed to identify vulnerabilities and strengthen software. But as these systems grow more capable, the industry will also face growing pressure to balance powerful offensive-style testing abilities with responsible safeguards.

TL;DR

• OpenAI has introduced GPT-5.5-Cyber, a new AI model in limited preview for defenders responsible for securing critical infrastructure.
• The launch puts OpenAI in more direct competition with Anthropic, which recently rolled out Claude Security and previewed its powerful Mythos model.
• Companies are increasingly using AI to find software vulnerabilities, suggest fixes, and speed up defensive cybersecurity tasks.
• Mozilla said Anthropic’s technology helped uncover 271 vulnerabilities during an early evaluation, showing how quickly AI tools are changing the security landscape.

This article was made with AI assistance and human editing.

Don’t forget to follow us for more exclusive content.

If you liked this, you might also like:

• The lawsuit of Zuckerberg against Zuckerberg
• Meta Faces a $109 Million Lawsuit From Eminem
• Microsoft faces a lawsuit for allegedly concealing a cheaper M365 subscription option