Cognition announced Devin Security 2.0 on October 26, describing managed scans that use frontier cybersecurity models, including Claude Mythos, alongside its model-agnostic Agentic MapReduce approach.
Security teams need more than a long list of potential vulnerabilities: they need evidence, prioritization and a path to a reviewed fix. Automated validation can help structure that work, but inconclusive outcomes and human review remain part of the process.
What happened?
Cognition announced Devin Security 2.0 on October 26, describing managed scans that use frontier cybersecurity models, including Claude Mythos, alongside its model-agnostic Agentic MapReduce approach.
The company says the service can prioritize findings in context and use a sandbox to validate selected high-priority issues. Its product page describes runtime validation as optional according to the scan profile and distinguishes confirmed results from false positives, findings requiring human review and inconclusive outcomes.
Cognition also describes remediation sessions that test fixes and open pull requests for a team to review. The company says findings remain open until a pull request merges and a recheck addresses the attack path. These are product claims, not an independent LessCarbs assessment.
Source: Cognition ↗ · Source: Cognition ↗
Why does it matter?
Security teams need more than a long list of potential vulnerabilities: they need evidence, prioritization and a path to a reviewed fix. Automated validation can help structure that work, but inconclusive outcomes and human review remain part of the process.
Read the sources
- Devin Security 2.0 launch ↗
Cognition · October 26, 2026 - Devin Security product page ↗
Cognition
