352/69 Tuesday, June 30, 2026
OpenAI has announced the launch of its latest AI models in the GPT-5.6 family as part of a limited testing release. The lineup consists of three model tiers based on use cases: Sol, Terra, and Luna. GPT-5.6 Sol is the flagship model with the highest capability and has been specifically designed for cybersecurity tasks. In the initial rollout, access will be limited to a group of trusted partners after the organization completed discussions with the U.S. government to establish a framework aligned with an executive order on assessing national security risks. This development represents an important step in applying advanced AI technology to strengthen defenses against cyber threats.
In terms of performance, testing in the ExploitBench environment found that GPT-5.6 Sol performed on par with a competing system, Mythos Preview, while using only one-third of the output tokens. The developers emphasized that Sol was designed primarily to support defensive missions, such as identifying vulnerabilities and developing security fixes, rather than enabling full-scale cyberattacks. Tests using the Chromium and Firefox browser source code showed that the system could distinguish software flaws and basic attack mechanisms, but could not independently generate a complete exploit chain. In addition, to control the risk of misuse, the system is governed by a multi-layered safety architecture that inspects input data in real time and uses more than 700,000 A100-equivalent GPU hours for automated red teaming to test and strengthen the system against attempts to bypass safety restrictions.
For system administrators and cybersecurity professionals, although the model is currently available only to a limited group through the API and Codex, OpenAI plans to expand access to ChatGPT and a broader user base in the coming weeks. Therefore, organizations and cybersecurity practitioners should closely monitor developments to study how this tool can be applied to code review and internal vulnerability assessment. However, when using AI, organizations should establish clear operational guidelines and data control policies to prevent sensitive information or confidential source code from being exposed to external systems during the review process.
Source: https://www.securityweek.com/openai-unveils-gpt-5-6-sol-as-its-most-advanced-cybersecurity-ai/