OpenAI introduced a derivative version of its latest AI model, GPT-5.5-Cyber, on Thursday. It is currently available in a limited preview to vetted cybersecurity teams. This move comes just one month after competitor Anthropic's preview of Claude Mythos attracted significant attention from investors and U.S. government officials.
In an official blog post, OpenAI stated that the GPT-5.5-Cyber preview is not intended to vastly enhance network capabilities. Instead, through targeted training, it has relaxed execution restrictions for security-related tasks. The company had released the general-purpose version of GPT-5.5 at the end of last month.
OpenAI claims this cybersecurity-specific model allows qualified teams to more easily utilize its latest large language model for workflows such as vulnerability identification and classification, patch effectiveness verification, and malware analysis. The built-in safety guardrails of the general-purpose GPT-5.5 significantly increase the operational difficulty for such specialized tasks.
OpenAI indicated in the blog post that GPT-5.5-Cyber is being opened to a small number of partners to facilitate their research into advanced security workflows, adapting to professional scenarios with specific requirements for exclusive access.
Last month, coinciding with the release of the Mythos model, Anthropic launched a specialized cybersecurity initiative called "Project Glasswing," granting model access to only a select few companies. Anthropic's CEO, Dario Amodei, met with senior officials from the Trump administration to discuss the model's technical capabilities and potential impact. This occurred even though the company had been added to the Pentagon's "supply chain risk" blacklist just weeks prior.
Reports indicate the U.S. Department of Defense requested that Anthropic allow the Pentagon to use its models without restriction for all lawful purposes. Anthropic, however, seeks to ensure its technology is not used for fully autonomous weapons or mass surveillance.
Early testing information about Mythos caused market volatility upon disclosure. In late March, as potential risks emerged—such as the model's ability to autonomously discover vulnerabilities and breach existing cyber defenses—market risk aversion quickly intensified. The cybersecurity sector faced collective pressure, with stocks like CrowdStrike, Palo Alto Networks, and Zscaler experiencing significant declines.
Mythos has also triggered high alert among U.S. regulators. Last month, Federal Reserve Chair Jerome Powell and Treasury Secretary Scott Besson reportedly held a closed-door meeting with Wall Street bank executives to discuss the potential cybersecurity threats posed by Mythos. Additionally, the U.S. Vice President Vance and Secretary Besson held advance phone calls with executives from major technology companies.
Notably, the competition between Anthropic and OpenAI has reached a fever pitch. Anthropic, originally focused on the enterprise AI sector, has recently made a strong push into the consumer market, with several practical products gaining rapid popularity. Leveraging a high-profile Super Bowl advertisement, it has positioned Claude as an ad-free alternative to ChatGPT, directly challenging OpenAI. Coupled with the publicity from its association with the Pentagon, Claude's ranking has soared, securing the second spot on the U.S. Apple App Store's free chart, positioned between the top-ranked ChatGPT and third-place Google Gemini, further intensifying its impact on OpenAI.
Comments