Key Highlights
- GPT-5.4-Cyber debuts exclusively for authenticated cybersecurity experts through OpenAI’s verification program
- The announcement arrives seven days following Anthropic’s introduction of Mythos, its security-centric AI system
- Federal Reserve Chairman Jerome Powell and Treasury Secretary Scott Bessent expressed alarm over Anthropic’s Mythos capabilities
- OpenAI intends to scale access from several hundred to multiple thousand authenticated security professionals soon
- Ramp payment analytics reveal approximately 33% of American companies subscribed to Anthropic’s services during March
OpenAI has unveiled a specialized artificial intelligence system engineered exclusively for cybersecurity applications. This new system, designated GPT-5.4-Cyber, is presently accessible to an exclusive circle of authenticated professionals.
The debut occurred on April 15, arriving precisely seven days after Anthropic introduced its restricted-access security model, Mythos.
GPT-5.4-Cyber functions as a “less restrictive” iteration of the standard GPT-5.4 framework. The system reduces rejection thresholds for cybersecurity operations that OpenAI classifies as authorized defensive activities.
Currently, availability remains confined to members of OpenAI’s Trusted Access for Cyber initiative. This program debuted in February and mandates identity authentication for professionals seeking model access.
OpenAI announced intentions to expand participation from several hundred current users to thousands of authenticated defenders over the forthcoming weeks. The organization emphasized that identity verification requirements will remain mandatory for all prospective users.
OpenAI initiated implementation of cybersecurity-specific protections across its model lineup throughout 2025. The company introduced Codex Security in March, designed to detect and remediate software security flaws.
Regulatory Alarm Over Anthropic’s Latest Model
Anthropic’s Mythos system has generated significant scrutiny from senior U.S. government officials. During the previous week, Federal Reserve Chair Jerome Powell and Treasury Secretary Scott Bessent cautioned financial industry executives to regard Mythos-related risks with utmost seriousness, Bloomberg reported.
Anthropic opted to distribute Mythos to approximately 40 major corporations to enhance their cybersecurity infrastructure. The restricted distribution strategy reflects apprehensions about potential misuse by malicious entities.
The director of Ireland’s National Cyber Security Centre testified before a parliamentary committee on April 14 that equivalent-capability models should be anticipated from threat actors before year’s end.
Anthropic co-founder Jack Clark projected that competing organizations will deploy comparable systems within the next several months. He further indicated that open-weight Chinese models featuring these capabilities might emerge within 12 to 18 months.
OpenAI’s Strategic Enterprise Focus
OpenAI has intensified its concentration on business clients, a sector where Anthropic has rapidly established market share. Analytics from payment processor Ramp indicate that roughly one-third of American enterprises purchased Anthropic solutions during March.
OpenAI has been scaling back initiatives with lower profit margins. The organization suspended development of an adult-oriented ChatGPT variant and halted its Stargate UK initiative.
Microsoft, serving as OpenAI’s primary investor, has committed to leasing data center infrastructure at a location initially designated for the Stargate Norway venture. A separate agreement with British AI infrastructure provider Nscale similarly collapsed.
OpenAI revealed plans to establish its inaugural permanent London headquarters in 2027, accommodating over 500 employees. The company characterized London as its most substantial research facility outside American borders.
OpenAI has additionally unveiled a new “superapp” concept following Anthropic’s lead, and introduced specialized healthcare AI capabilities shortly after Anthropic deployed its health-oriented Claude enhancements.
