After Anthropic Mythos: OpenAI announces GPT-5.4-Cyber
OpenAI is launching its own AI model for cybersecurity, GPT-5.4-Cyber. As with Anthropic's Mythos, access will initially be restricted.
Anthropic's Mythos has caused a stir in the tech world and politics. Now comes OpenAI's response.
(Image: Prathmesh T/Shutterstock.com)
After Anthropic's AI model Mythos caused a stir in the highest US government circles recently causing a stir in the highest US government circles, OpenAI is now introducing its model tailored for defensive security tasks: GPT-5.4-Cyber.
Access is restricted and is provided through the Trusted Access for Cyber (TAC) program launched in February. OpenAI plans to expand it in the coming weeks to thousands of verified security researchers and hundreds of teams responsible for protecting critical software. The program will also be expanded with additional access levels. At the highest levels, GPT-5.4-Cyber will be unlocked, a specially adapted variant with fewer security restrictions and extended functions, including reverse-engineering capabilities that enable the analysis of finished programs without source code, writes OpenAI.
The introduction will be gradual, and restrictions may apply depending on the use case. Access to TAC is via identity verification for individuals or upon request for companies.
In addition, OpenAI refers to Codex Security, a security agent tested for months that automatically monitors codebases, identifies vulnerabilities, and suggests repairs. Since its launch as a research preview, the system is said to have contributed to more than 3,000 critical and serious security vulnerabilities being fixed.
AI Models Increase Cybersecurity Risks
OpenAI sees GPT-5.4-Cyber as a first step towards further AI models specifically tailored for cybersecurity. The company considers the current protective mechanisms sufficient to make current models more broadly accessible, while variants specifically loosened for cybersecurity work should initially remain under stricter controls. In the long term, OpenAI expects that significantly more extensive protective measures will be necessary for even more powerful models.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmung wird hier ein externes YouTube-Video (Google Ireland Limited) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (Google Ireland Limited) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
The announcement follows a week after the presentation of Anthropic's AI model Mythos, which the company is initially not releasing broadly due to its high potential for misuse. According to Anthropic, the model has already identified thousands of zero-day vulnerabilities in major operating systems, browsers, and other critical software. Through the “Project Glasswing” initiative, only selected companies and organizations will initially have access to secure critical software before comparable capabilities become more widely available. Participants include Apple, Google, Microsoft, and Nvidia. OpenAI does not name any partners in its announcement.
If the capabilities attributed to Mythos are confirmed, such AI models could fundamentally change the rules of cybersecurity, with consequences far beyond the IT industry. The fact that the US Treasury Department, the Federal Reserve, and major banks have already been alerted underscores the potential impact on critical infrastructure and financial stability.
(mma)
