Quick Summary
- GPT-5.4-Cyber became available to a curated group of authenticated cybersecurity experts via OpenAI’s Trusted Access program
- The announcement arrived seven days following Anthropic’s deployment of Mythos, its security-oriented AI system
- Federal officials including Treasury Secretary Scott Bessent and Fed Chair Jerome Powell expressed caution regarding Anthropic’s Mythos capabilities
- OpenAI will scale access from several hundred to thousands of authenticated security professionals over upcoming weeks
- Ramp’s payment data revealed approximately 33% of American enterprises subscribed to Anthropic services during March
OpenAI unveiled a specialized AI system designed exclusively for cybersecurity applications. The platform, named GPT-5.4-Cyber, launched with restricted availability to authenticated professionals.
The deployment occurred on Tuesday, April 15. Anthropic had introduced its security-focused system, Mythos, to select organizations exactly seven days earlier.
GPT-5.4-Cyber operates as an enhanced variant of GPT-5.4 with adjusted parameters. The system reduces restriction thresholds for cybersecurity operations that OpenAI classifies as authorized defensive activities.
Current availability extends exclusively to members of OpenAI’s Trusted Access for Cyber initiative. This program began operations in February and mandates identity authentication for all participants seeking model interaction.
OpenAI outlined expansion plans to scale the initiative from hundreds of current users to thousands of authenticated defenders throughout coming weeks. The organization emphasized its commitment to maintaining identity verification protocols before approving access.
OpenAI began implementing cybersecurity-specific protective measures across its model lineup throughout 2025. The company introduced Codex Security in March to assist organizations in detecting and resolving software security flaws.
Federal Response to Anthropic’s Mythos
Anthropic’s Mythos platform attracted scrutiny from senior U.S. government officials. During the previous week, Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell advised financial industry executives to recognize potential threats associated with Mythos, per Bloomberg reporting.
Anthropic made a strategic decision to distribute Mythos to approximately 40 major corporations for strengthening their cybersecurity infrastructure. The company implemented strict access controls stemming from concerns about unauthorized acquisition.
Ireland’s National Cyber Security Centre director addressed a parliamentary committee on April 14, projecting that threat actors would develop equivalent-capability models before year-end.
Anthropic co-founder Jack Clark predicted competing organizations would deploy comparable systems within several months. He projected that open-weight Chinese models featuring these capabilities could emerge within 12 to 18 months.
OpenAI’s Corporate Market Strategy
OpenAI has intensified efforts targeting enterprise clientele, a segment where Anthropic has achieved rapid adoption. Payment processing firm Ramp’s data indicates roughly 33% of American businesses subscribed to Anthropic’s offerings during March.
OpenAI has discontinued several lower-margin initiatives. The organization suspended development of an adult-oriented ChatGPT variant and halted its Stargate UK infrastructure project.
Microsoft, serving as OpenAI’s primary financial supporter, secured an agreement to lease data center facilities at a location initially designated for the Stargate Norway development. A separate partnership arrangement with UK AI infrastructure provider Nscale collapsed.
OpenAI revealed plans to establish its first permanent London headquarters in 2027, featuring capacity for over 500 employees. The company identified London as its largest research facility outside American borders.
OpenAI recently introduced a comprehensive “superapp” following Anthropic’s example, and rolled out specialized healthcare AI capabilities shortly after Anthropic deployed its health-oriented Claude functionalities.
