Cognition Announces 'Devin Security Swarm' to Help Resolve Vulnerability Backlogs

San Francisco — June 30, 2026 — Cognition (Headquarters: California, USA; CEO and Co-Founder: Scott Wu), the global leader in autonomous AI agents for software development, today announced 'Devin Security Swarm,' designed to assist security teams by detecting exploitable vulnerabilities, validating them during runtime, and repairing them at lower cost than similar solutions.

Devin Security Swarm was developed in response to an era where AI agents generate code faster than security teams can review it. Security issues detected per company have increased from approximately 1,000 to over 10,000 within six months, driven by the fact that 42% of all code is now generated or assisted by AI. Enterprises releasing more code than ever before now need to strengthen their security teams to keep pace with these changes.

Devin Security Swarm helps security teams respond to vulnerabilities at the same speed and scale as engineering teams. Built on Devin, it verifies whether detected vulnerabilities are actually exploitable and automatically generates pull requests with fixes directly within the development workflows engineers use daily. This enables security teams to respond quickly to newly discovered vulnerabilities and prevents backlogs from growing faster than their capacity to respond.

Nick Wong, Cognition's Lead for Security Engineering, stated:

'Devin Security Swarm delivers engineering-team-level responsiveness that security teams have never had before. This allows security teams to directly fix vulnerabilities instead of just verifying exploitability and waiting for engineering teams to act.'

Performance Based on Verified Data

For enterprise security teams, it's critical that tools perform well at scale and prove their findings represent real vulnerabilities that need addressing. To answer these questions, Cognition demonstrates Security Swarm's performance using evidence from public benchmarks.

● Comprehensive Detection for Large-Scale Environments:

Devin Security Swarm uses an agent-based MapReduce architecture, where multiple AI agents analyze and reason across files to detect vulnerabilities hidden in real application behavior—such as business logic flaws and authentication bypasses across services. Detected vulnerabilities are then reproduced in isolated sandbox environments to verify actual exploitability during runtime. For vulnerabilities confirmed as exploitable, Devin generates patches and automatically creates pull requests with the fixes.

● Proven High Performance and Superior Cost Efficiency:

In a benchmark across 14 programming languages targeting 50 real-world vulnerabilities published in GitHub Security Advisories, Devin Security Swarm detected 36 vulnerabilities. This exceeds the detection count of other AI-based vulnerability scanners tested and achieved a 30% lower cost per detection compared to the most comparable high-performing service. Furthermore, Devin Security Swarm uniquely detected 3 critical vulnerabilities that no other tool identified.

Security Vulnerability Response Program

To help enterprises enhance their security posture, Cognition offers a 6-week structured program called the 'Devin Security Program.' This program assesses an organization's application security posture and helps resolve existing vulnerability backlogs. Through this program, security leaders can systematically eliminate backlogs and implement AI agents to maintain codebase health and continuously detect vulnerabilities.

For details on the Devin Security Program, visit here (in English).

Devin Security Swarm is available globally today for enterprise customers. Learn more here (in English).

About Cognition

Cognition is an AI lab building the future of software, and the creator of the AI software engineer 'Devin.' Devin performs complex engineering tasks end-to-end—from code design and implementation to testing and fixing security vulnerabilities—enabling teams to focus on more creative challenges. Visit Cognition.com for more information.