SAN FRANCISCO — June 30, 2026 — Cognition, a global leader in autonomous AI agents for software development, today announced Devin Security Swarm, which assists security teams in detecting exploitable vulnerabilities, validating them at runtime, and remediating them at a lower cost than similar solutions.
Devin Security Swarm was developed against a backdrop where AI agents are now generating code faster than security teams can review it. The number of security issues detected per company per month has increased from approximately 1,000 to over 10,000 in six months, partly due to AI generating or assisting in the creation of 42% of all code. Companies releasing more code than ever before are being challenged to scale their security teams to keep pace with these changes.
Devin Security Swarm empowers security teams to respond to vulnerabilities at the same speed and scale as engineering teams. Built on Devin, it validates whether detected vulnerabilities are actually exploitable and automatically generates the necessary pull requests for remediation within the development workflows that engineering teams use daily. This allows security teams to quickly address newly discovered vulnerabilities and prevents backlogs from growing faster than teams can handle them.
"Devin Security Swarm brings engineering-level responsiveness to security teams, something they haven't had before," said Nick Wong, Security Engineering Lead at Cognition. "This allows security teams to validate if a vulnerability is actually exploitable and directly remediate it, rather than waiting for engineering teams to do it."
Performance Backed by Proof
For enterprise security teams, it is critical that tools perform at scale and can prove that their findings represent actual vulnerabilities that need addressing. Cognition has demonstrated Security Swarm's performance with evidence from public benchmarks to answer these questions.
● Comprehensive Detection for Large-Scale Environments:
Devin Security Swarm employs an agent-based MapReduce architecture, where multiple AI agents analyze and reason across files to detect vulnerabilities lurking in the application's actual behavior, such as flaws in business logic or authentication bypasses across services. Furthermore, it reproduces detected vulnerabilities in isolated sandbox environments to validate that they are indeed exploitable at runtime. For vulnerabilities deemed exploitable, Devin generates patches and automatically creates pull requests reflecting the fixes.
● Proven High Performance and Superior Cost Efficiency:
In a benchmark conducted across 14 programming languages on 50 real vulnerabilities published on GitHub Security Advisories, Devin Security Swarm detected 36 vulnerabilities. This detection count surpassed other AI-based vulnerability scanners tested, and the cost per detection was 30% lower than the closest comparable service. Moreover, Devin Security Swarm was the only tool to detect three critical vulnerabilities that all other tools missed.
Security Vulnerability Response Program
To support enterprises in advancing their security posture, Cognition offers a six-week, structured program called the "Devin Security Program." This program assesses an organization's application security posture and helps clear existing vulnerability backlogs. With this program, security leaders can systematically clear their vulnerability backlogs and implement AI agents that continuously maintain code base health and a mechanism for ongoing vulnerability detection.
Learn more about the Devin Security Program here.
Devin Security Swarm is available globally to enterprise customers starting today. Learn more here.
About Cognition
Cognition is an AI lab building the future of software and the creator of Devin, an AI software engineer. Devin performs complex engineering tasks end-to-end, from designing and implementing code to testing and fixing security vulnerabilities, enabling teams to focus on more creative challenges. Learn more at Cognition.com.
FACT BOX
- Source: PR TIMES
- Category: 製品発表
- Organizations: Cognition / GitHub