· Updated
GPT-5.4-Cyber, a version of the GPT-5.4 frontier model fine-tuned (adapting a pre-trained model for specific tasks) to be cyber-permissive for defensive use cases.As models reach high-capability thresholds, general-purpose safeguards often block legitimate security work like vulnerability research. By shifting to an identity-based access model, defenders can use powerful reasoning without constant refusals. This strategy ensures that security professionals have tools that scale in lockstep with the offensive potential of AI.
You can now access advanced capabilities like binary reverse engineering, which allows for analyzing compiled software for malware or vulnerabilities without source code. Individual defenders can verify their identity through a dedicated portal, while enterprises can request access through representatives. Approved users gain access to models with reduced friction for dual-use activity.
We’re expanding Trusted Access for Cyber with additional tiers for authenticated cybersecurity defenders. Customers in the highest tiers can request access to GPT-5.4-Cyber, a version of GPT-5.4 fine-tuned for cybersecurity use cases, enabling more advanced defensive workflows. https://t.co/RMMXQklFar
More like this
OpenAIMar 5
OpenRouterApr 24
OpenClawApr 15