© ROOT-NATION.com - Use of content is permitted with a backlink.
OpenAI has unveiled GPT-5.4-Cyber, a specialized version of its flagship AI adapted for defense cybersecurity tasks. This happened a week after its competitor, Anthropic, announced the launch of its own advanced model, Claude Mythos.
The Mythos model, announced on April 7, is being deployed as part of Anthropic’s Project Glasswing initiative. As part of this controlled project, a limited number of organizations are given access to the Claude Mythos Preview version for use in defensive cybersecurity. The system has reportedly helped identify thousands of serious vulnerabilities in operating systems, web browsers, and other software.
Reas also: Claude Mythos: Too Intelligent for Its Users?
OpenAI, the developer of the popular ChatGPT chatbot, noted that GPT-5.4-Cyber will also be available at the first stage to a limited number of trusted security solution providers, organizations, and researchers. This approach is explained by the model’s more flexible architecture, which provides wider capabilities in dealing with sensitive tasks.
In addition, the company announced the expansion of its Trusted Access for Cyber program. From now on, it will cover thousands of trusted individual cybersecurity professionals and hundreds of teams involved in critical software security. OpenAI is also expanding the TAC program with new access levels. It was launched in February and now provides an expanded verification system, where higher levels provide access to more powerful features.
Read also: Trump vs. Claude AI: How the AI Debate Is Unfolding in the U.S.
The highest levels provide access to GPT-5.4-Cyber, a model specifically customized for advanced cyber capabilities with fewer restrictions. It lowers the failure threshold for legitimate tasks and supports complex defense scenarios, including analyzing compiled software without access to the source code to identify malicious components and vulnerabilities. Access to such tools may be accompanied by additional restrictions, in particular in cases where there is insufficient visibility of use.
“Our cybersecurity defenses are the result of many months of iterative improvement. We believe the class of safeguards in use today sufficiently reduce cyber risk enough to support broad deployment of current models” OpenAI said in a statement. “We expect versions of these safeguards to be sufficient for upcoming more powerful models, while models explicitly trained and made more permissive for cybersecurity work require more restrictive deployments and appropriate controls”.
Read also:
- BeingBeyond is revolutionizing the industry: The world’s most powerful AI model is unveiled
- Source Code for Anthropic’s Claude Code Leaks at the Exact Wrong Time