OpenCSG releases Agentic-27B model for tool-using workflows requiring commercial permission

OpenAI has launched GPT-6 Cyber, an artificial intelligence model focused on cybersecurity, which demonstrated the ability to identify and validate thousands of exploits in programs such as Linux and Firefox. Furthermore, the tool assists in developing methods to exploit vulnerabilities and suggests corrections.

Access to this technology will be limited to professionals and organizations approved by the company itself through the security program called Daybreak.

In the ExploitGym test, GPT-6 Cyber achieved an accuracy of 48.5%, representing a relative improvement of 51% compared to GPT-5.6 Cyber, which scored 32.1%. In another evaluation, focused on detecting flaws in web applications without access to the source code, the new model managed to identify 40.3% of vulnerabilities, surpassing the 32% recorded by GPT-6 Astra.

OpenAI also verified whether the tool respected the limits established by researchers. In 99.93% of tests, GPT-6 Cyber avoided exploiting services outside the authorized scope without receiving additional security instructions, and this rate rose to 100% when deployment instructions were provided.

The company classified the model as Critical in cybersecurity, a category higher than the High classification given to GPT-5.6 Cyber within its risk assessment framework.

The model's results were also observed in open-source projects. In the Linux system, GPT-6 Cyber reproduced 1,724 vulnerabilities and created functional exploits for 948 of them. This number is more than five times greater than the 166 cases exploited by GPT-5.6 Sol using the same research procedure.

OpenAI notified project managers about 304 critical findings, providing suggested corrections, while the remaining vulnerabilities are in the correction phase. In Firefox, the model identified 81 novel flaws, which were classified as high severity by Mozilla. Additionally, the work revealed that another 30 known issues were more serious than previously thought. OpenAI delivered reproducible tests and repair proposals to the browser team, which has already begun releasing updates to fix some of these flaws.

GPT-6 Cyber is integrated into Daybreak, an OpenAI program dedicated to digital security defense and research. This program has three access modalities. Initially, GPT-6 Cyber will be accessible via API and Codex for a specific group of approved Daybreak Red clients, with plans for gradual expansion to other participants in this modality.

The company also announced a $1 billion subsidy program to facilitate the use of Daybreak tools by under-resourced teams, covering government agencies and essential service providers.

To mitigate the risk of misuse, OpenAI implemented monitoring of suspicious activities, restrictions on the use of compromised credentials, and mechanisms to prevent assistance in unauthorized attacks against vital systems. The company also advises the use of isolated environments and human supervision in high-risk operations.

OpenAI professionals signaled to management for a long period vulnerabilities in the security of the company's internal systems. As reported by Reuters, these flaws could allow external individuals, including foreign governments, to gain access to crucial data about artificial intelligence (AI) progress.

The complaints were brought to the attention of executives and the board. Some collaborators concluded that the organization prioritized development speed to compete with giants like Google and Anthropic, neglecting security.

Five people familiar with OpenAI's practices reported that employees with code access could view details of various projects. This occurred because a large part of the code was concentrated in a single repository, known as a monorepo.

OpenAI itself stated that it has implemented restrictions on access to sensitive work and that vital parts of its research are kept in separate repositories. The company clarified that the main repository does not contain all the code or model weights.

Concerns also covered the absence of periodic security audits. Three sources indicated that modifications made by employees were not always subjected to checks through internal tools.

One specific incident illustrated another facet of the risks. On October 13, the U.S. Department of Justice announced the arrest of Abdulrahman Alfozan, an OpenAI engineer, accused of hiding a recorder in a toilet paper roll in the executive meeting room.

Court documents revealed that he used an internal AI tool to track information about executive travel to Saudi Arabia and potential meetings with ministers. Alfozan was in contact with a member of the Saudi royal family who had tried to recruit him.

In response to concerns regarding system protection, OpenAI has also strengthened its security infrastructure. However, the issues of safeguarding strategic information affect other AI companies as well.

In January, a federal jury sentenced a former Google employee for stealing the company's technological secrets for China. At the end of October, the startup Cathedral Systems claimed to have hacked into Anthropic's systems and accessed Slack conversations about models, obtaining data such as client and partner contacts, as well as basic account records.

Such occurrences intensify the debate about the open research culture in the sector. At OpenAI, employees have begun to question whether information sharing is compatible with protecting strategic technologies.

In 2023, Leopold Aschenbrenner, then a researcher at the company, wrote a memo about the danger of Chinese infiltration. This document was presented to the board after an external intruder gained details about GPT-4. OpenAI informed that the intruder only accessed Slack channels open to the entire team.

This year, OpenAI's AI agents in tests infiltrated Hugging Face systems, an event analyzed by US authorities. In July, researchers from Hacktron AI used an Anthropic model to penetrate OpenAI employees' ChatGPT accounts and access internal systems. OpenAI thanked the warning and announced that it had adopted measures to reduce the risk of similar attacks.

Former researcher Will DePue, who left the company in April, criticized the emphasis on speed. In a post on X, he argued that OpenAI considered advancing quickly more important than protecting its systems, even against major national adversaries.

Spokesperson Elana Widmann told Reuters that the company invests in defense against internal threats, cybercriminals, and governments. After being contacted by the agency, OpenAI disclosed recent security improvements. According to the report, this strengthening of protections may slow down AI advancement and complicate the IPO plans of OpenAI and Anthropic.

The OpenAnolis community is developing the next version of its Linux distribution, Anolis OS 27, using an approach they call agentic. In this approach, various artificial agents take on functions such as package selection and fixes, code verification, and patch backporting.

Ma Tao, chairman of the OpenAnolis community and head of the base software department at Alibaba Cloud, presented this plan at the GOSIM AI Vision Forum in Zhuhai this month. He also shared data from his team, which indicates that 95 percent of new code in the Alibaba Cloud base software team is generated by agents.

The goal stated for OpenAnolis is to accelerate engineering processes throughout the entire operating system lifecycle—from development and maintenance to operation and management—by tenfold. Ma Tao noted that agents could potentially create a complete operating system independently in the future, and the industry must embrace this trend. Both figures relate to Alibaba Cloud.

Ma Tao spoke at the forum's opening roundtable titled 'Roots: Open Source as the Foundation in the Agent Era.' The discussion was moderated by the co-chair of the LF AI and Data Generative AI Commons, and participants included leaders from the Rust Foundation, OpenUK, OpenCV, and the Beijing Academy of Artificial Intelligence. The panel discussed issues of governance, code accountability, and global collaboration as AI begins to provide and verify code, changing the composition of open-source project contributors.

Ma Tao described two areas of work. The first is the creation of a base operating system for agents. In this concept, the OS goes beyond being a simple resource manager; it becomes an infrastructure that connects computing power, models, tools, and applications. According to data presented at the Apsara 2026 conference, Alibaba Cloud's Agentic OS reduces token usage by over 30 percent and doubles sandbox deployment density through system-level configuration for short-term agent sandboxes and frequent process creation. OpenAnolis's ANOLISA project is positioned as the runtime environment for each agent.

The second area is Anolis OS 27 itself. The community has made the Anolis OS 27 monorepository publicly available, which is a single Git repository containing a large number of packages. OpenAnolis describes this as a distribution design method created for collaboration between human maintainers and AI agents, allowing other teams to study and utilize it.

Anolis OS is developed in collaboration with companies, universities, and research institutes, and it serves as a higher-level community for Alibaba Cloud Linux.