Cogent's VR-1.1 Leads in LiveCVEBench Vulnerability Fix Test

icon币界网
Share
AI summary iconSummary
Cogent Security's VR-1.1 leads in LiveCVEBench vulnerability news, achieving a 57.9% fix rate on 190 recent CVEs. The model outperformed Claude Code and Codex, both at 45.3%. LiveCVEBench, developed by Harbin Institute of Technology and ByteDance, evaluates real-world code fixes. VR-1.1 also generates regression tests to prevent vulnerabilities from re-emerging. Cogent aims to accelerate responses to security breach risks.
CoinDesk reports:

Cogent releases its latest cutting-edge cybersecurity model, VR-1.1, ranking first in the LiveCVEBench vulnerability remediation test.

Among the 190 recently disclosed CVEs, VR-1.1 achieved a 57.9% effective fix rate; Claude Code and Codex both achieved 45.3%.

San Francisco, November 3, 2026 /PRNewswire/ -- Cogent Security today released VR-1.1, the latest version of its cutting-edge reasoning model for cybersecurity. VR-1.1 ranks among the top performers on LiveCVEBench, a public benchmark that tests AI agents’ ability to fix recently disclosed vulnerabilities in real codebases. Cogent reports that VR-1.1 successfully fixed 57.9% of the 190 tasks on this benchmark, compared to 45.3% for Claude Code and Codex.

The number of vulnerabilities known to security teams far exceeds what they can fix, and AI has significantly accelerated the rate at which new vulnerabilities are discovered. Cogent reports that in July, the number of high-severity and critical vulnerabilities disclosed by major software vendors reached five times the previous monthly record set in April, before Anthropic announced the Claude Mythos Preview. Each fix still relies on engineers first fully understanding the vulnerability before modifying code without disrupting other components—a process that determines the pace of most vulnerability management initiatives.

Vineet Edupuganti, Co-founder and CEO of Cogent Security, said: “The speed at which vulnerabilities are being discovered this year has increased significantly. Fixing them, however, remains slow—and that’s where security teams are getting stuck. Every company will need AI that can fix issues as quickly as AI discovers them, without breaking the systems their business depends on. We built VR-1.1 for this purpose; on a public benchmark, it fixes more real vulnerabilities than leading general-purpose AI programming tools.”

Rated by security and functionality

LiveCVEBench was created by researchers from Harbin Institute of Technology and Kuaishou Technology. It transforms newly disclosed CVEs into repair tasks across 14 programming languages, covering vulnerabilities such as injection, access control, configuration, and memory safety, and regularly incorporates new CVEs. Each task places an agent in a real development environment, providing both the vulnerability report and the affected codebase.

A patch is considered passed only when the baseline security tests show that vulnerabilities have been closed and functional tests confirm that the software continues to operate as expected. Cogent states that, in its evaluation, the remediation phase blocked external network access and scoring tests were not provided until each proxy was completed. The reasoning budgets for the three systems were not identical.

How to complete the repair for VR-1.1

Where environment permits, VR-1.1 will first write its own proof-of-concept to trigger the vulnerability reported. It will trace the propagation path of the vulnerability within the affected code, including related paths sharing the same weakness, and apply targeted fixes—such as checking indices before use. A subsequent review will then examine the patch in a new context, seeking ways to bypass the fix and determining whether the modification introduced new vulnerabilities.

VR-1.1 will also write regression tests to catch the vulnerability if it reappears due to future code changes. If an attempt fails, it will carry forward the previous diagnostics and failure details into the next attempt.

Geng Sng, Co-founder and CTO of Cogent Security, said: “A patch might block the exact payload mentioned in the vulnerability report, but the same vulnerability could still be exposed through another endpoint or disrupt core user functionality. We designed VR-1.1 to check for both types of issues simultaneously before completion. The security team receives a patch ready for review, along with diagnostics and test results.”

VR-1.1 extends the remediation capabilities introduced by VR-1, which Cogent launched in July. VR-1 is the company’s cutting-edge reasoning model designed to identify and prove potential attack paths an attacker could exploit within an enterprise. Cogent Research’s full documentation, including methodology, is available here.

About Cogent Security

Cogent Security is an AI lab building an enterprise platform for autonomous cyber defense. Its AI agents discover and remediate vulnerabilities at machine speed, enabling every company to run its own Cyber Defense Factory. Starting with autonomous vulnerability management, Cogent guides vulnerabilities from disclosure to verified remediation across cloud, on-premises, and endpoint environments—identifying exposure surfaces in minutes and confirming exactly where attackers can reach. Customers define the policies their agents follow and determine which remediations can be executed autonomously and which require approval. Security teams at Fortune 500 companies using Cogent have reduced the exposure window for critical vulnerabilities by 97%. Cogent was founded by researchers and operators from Google DeepMind, Abnormal Security, and Coinbase, and is backed by Greylock Partners and Bain Capital Ventures. Learn more at cogent.com.

SOURCE Cogent Security

Disclaimer: The information on this page may have been obtained from third parties and does not necessarily reflect the views or opinions of KuCoin. This content is provided for general informational purposes only, without any representation or warranty of any kind, nor shall it be construed as financial or investment advice. KuCoin shall not be liable for any errors or omissions, or for any outcomes resulting from the use of this information. Investments in digital assets can be risky. Please carefully evaluate the risks of a product and your risk tolerance based on your own financial circumstances. For more information, please refer to our Terms of Use and Risk Disclosure.