OpenAI AI Pipeline generates 948 Linux root escalation exploits in 30 hours of execution
OpenAI on Wednesday detailed a pipeline that used an upcoming cybersecurity-focused artificial intelligence (AI) model to validate and remediate 948 Local Elevation of Privilege Vulnerabilities in Linux for $1.2 million in API costs.
“Our prototype identified and analyzed 2,308 candidate Linux kernel hits, dynamically reproduced 1,724 vulnerabilities, and produced 948 local privilege escalation (LPE) exploits that could allow an unprivileged local user to gain root access,” the AI company said.
The results build on 24 LPE exploits revealed by OpenAI and discovered by its GPT-5.5-Cyber model in June 2026. The latest research transforms this process into an agent pipeline comprising five stages:
- Vulnerability discovery, where agents identify high-value audit targets (i.e., security-relevant parts of the code base that are accessible by low-privileged users) and search for candidate results.
- Dynamic validation, where agents attempt to reproduce each candidate in a sandbox environment and preserve execution proofs
- Exploitability assessment, where agents test whether validated vulnerabilities could be exploited to facilitate information disclosure or LPE and design a proof-of-concept (PoC) exploit.
- Patch generation, where agents generate patches for the identified vulnerability
- Human review, where security researchers review the exploit, fix, and proposed severity before they are reported to maintainers.
“During a mid-October 2026 run, our pipeline using a future cyber model dynamically reproduced 1,724 vulnerabilities (in 30 hours) and generated PoCs for 291 information leak vulnerabilities and 948 working local privilege escalation (LPE) exploits for a commonly used Linux distribution,” OpenAI said.
“The total run cost $1,213,430 in API credits, or $1,279 per LPE exploit. To our knowledge, agents in this pipeline have discovered and used at least 10 new Linux exploitation techniques.
At a high level, these include:
- Two new persistent elastic objects to spray objects in the target kernel cache
- Two new page cleanup techniques to improve exploit reliability
- Five new data-only techniques for achieving privilege escalation directly or via arbitrary write
- A technique that involves arbitrary type confusion
This development is once again an indication of how cutting-edge AI is reshaping the contours of cybersecurity, accelerating the discovery and exploitation of vulnerabilities, and at the same time helping defenders reliably fix flaws, albeit through additional instructions and iterations, before attackers can use them, and making it “prohibitive” for bad actors to exploit software.
In a report released last month, VulnCheck classified AI-assisted vulnerability discovery as a “growing risk category” that is boosting entry-level hackers, who now require little or no technical expertise to detect software vulnerabilities.
Access to AI models is not only limited to defenders, but also to bad actors who can exploit the same capabilities to discover and weaponize new zero-days and expand the pool of viable targets with minimal effort.
According to information from the Google Threat Intelligence Group (GTIG), the number of vulnerabilities disclosed each month more than doubled, from 5,045 in January 2026 to 10,740 in August 2026. The number of vulnerabilities classified as high risk increased from 131 disclosures in January 2026 to 350 in August 2026, a growth of 167%.
“The widespread availability of AI-based cyber tools has almost certainly led to a significant increase in the discovery of zero-day vulnerabilities,” the UK’s National Cyber Security Center (NCSC) said in an assessment published on Tuesday.
“AI defensive cyber tools have reduced the window for going from a zero-day vulnerability to an N-day vulnerability, and accelerated the patching process. However, without proper and widespread configuration of this technology, many systems remain vulnerable to attackers using cyber AI tools to identify and exploit system weaknesses.
The NCSC also highlighted that organizations will need to evolve their internal patching and testing processes to meet this new pace of vulnerability discovery.
“Attackers have subsequently become less cautious about losing access to achieve an objective because the replacement cost is minimal,” he adds. “Cyber capabilities are now used for a wider range of activities due to the lower risk involved, making it easier for adversaries to access information and use it operationally.”
OpenAI said the rate of vulnerability discovery by frontier models must be matched by automated patch generation to secure software and limit the exploitation window.
“Through human expert-guided iteration, we believe AI-generated patches can be functionally correct, aligned with existing coding styles, and robust against attack variants,” OpenAI said. “In the short to medium term, access to a low-privileged account on an operating system that is not up to date with its latest available patches should be considered functionally equivalent to root access for a well-resourced threat actor.”
Gn tech
