research
Building frontier AI for autonomous cyber defense
Cogent Research works on the hardest open problems in AI × Security with access to real enterprise environments that no pure research setting can provide.
Introducing Cogent Research
Watch Video
Introducing Cogent Research
Leading AI and Security researchers
Former Google DeepMind researchers advancing frontier AI systems for security-critical environments.
Leading AI and Security researchers
Former Google DeepMind researchers advancing frontier AI systems for security-critical environments.
Deep peer-reviewed research foundation
The team has 8,000+ citations across AI, systems, and security research, from work published in leading venues.
Deep peer-reviewed research foundation
The team has 8,000+ citations across AI, systems, and security research, from work published in leading venues.
World-class security expertise
Team responsible for security systems that protect more than half of the Fortune 500.
World-class security expertise
Team responsible for security systems that protect more than half of the Fortune 500.
W9h%yE C9oJg3eLn5tE
WHhBy4 CYoHg0eKn&tS
WKh0y8 CUo7g7e%nRtM
The missing environment for AI security research
The hardest problems in autonomous security cannot be solved in isolation. They require frontier AI talent, real enterprise data, security expertise, production feedback loops, and rigorous safety constraints. Historically, those ingredients have lived in separate worlds.
AI Labs
World-class model research, infrastructure, and compute. Limited access to real enterprise security environments, operator workflows, and remediation outcomes.
AI Labs
World-class model research, infrastructure, and compute. Limited access to real enterprise security environments, operator workflows, and remediation outcomes.
Cogent
Frontier AI researchers working with real security data and production systems to advance the field while solving one of the most important applied problems today.
Cogent
Frontier AI researchers working with real security data and production systems to advance the field while solving one of the most important applied problems today.
Security Vendors
Deep domain expertise and large operational datasets. Limited ability to conduct frontier AI research in agents, reasoning, formal methods, and model development.
Security Vendors
Deep domain expertise and large operational datasets. Limited ability to conduct frontier AI research in agents, reasoning, formal methods, and model development.
TGh3eI TCe2a@m&
TTh6eT T9e3aDm7
TChHe8 TIe6aFm7
The researchers behind Cogent
Duc Hiep Chu
Head of Research
Former Google DeepMind researcher with 4,000+ academic citations. Co-author of CodeMender, an AI agent for identifying and repairing code vulnerabilities before they propagate through downstream systems. At Cogent, Duc leads research into autonomous remediation, formal validation, and the systems required for safe AI-driven security action.
Duc Hiep Chu
Head of Research
Former Google DeepMind researcher with 4,000+ academic citations. Co-author of CodeMender, an AI agent for identifying and repairing code vulnerabilities before they propagate through downstream systems. At Cogent, Duc leads research into autonomous remediation, formal validation, and the systems required for safe AI-driven security action.
Anirudh Ravula
Head of AI
Former Google DeepMind researcher with 4,000+ academic citations. Led teams working on transformer architectures and large language models. At Cogent, Anirudh leads the development of security-specialized reasoning systems: models that can understand complex environments, evaluate risk, and support reliable autonomous defense.
Anirudh Ravula
Head of AI
Former Google DeepMind researcher with 4,000+ academic citations. Led teams working on transformer architectures and large language models. At Cogent, Anirudh leads the development of security-specialized reasoning systems: models that can understand complex environments, evaluate risk, and support reliable autonomous defense.
Geng Sng
Co-Founder & CTO
Built one of the world’s leading ML fraud detection systems at Abnormal Security, with patented technology used by half of the Fortune 500. At Cogent, Geng leads the platform architecture that connects frontier research to production security systems: real-time data, enterprise-scale reasoning, and operational feedback loops.
Geng Sng
Co-Founder & CTO
Built one of the world’s leading ML fraud detection systems at Abnormal Security, with patented technology used by half of the Fortune 500. At Cogent, Geng leads the platform architecture that connects frontier research to production security systems: real-time data, enterprise-scale reasoning, and operational feedback loops.
RBe6sEe%aZr1c3h2 P@rZiMo#rCiCt5iPe#sX
RTe2sYeXaUrCc5h9 PVrNiRo%r4iKt#i4e#s9
RJeBsXe2aNrWcAh9 PFrFiMo$rAi5tMiUeLsP
What we’re working on
Autonomous Security Agents
Agents capable of long-horizon defensive security tasks across complex enterprise environments — with policy constraints, auditable reasoning, escalation paths, and rollback.
Autonomous Security Agents
Agents capable of long-horizon defensive security tasks across complex enterprise environments — with policy constraints, auditable reasoning, escalation paths, and rollback.
Formal Security Proof Systems
Proof systems, solvers, and validation frameworks that generate machine-checkable evidence that a remediation is safe, correct, and does not introduce new risk.
Formal Security Proof Systems
Proof systems, solvers, and validation frameworks that generate machine-checkable evidence that a remediation is safe, correct, and does not introduce new risk.
Security Benchmarking
Benchmarks, safety frameworks, open-source tools, and evaluation methods for measuring autonomous security systems under real-world conditions: noisy data, adversarial ambiguity, cost-quality tradeoffs, and safe action.
Security Benchmarking
Benchmarks, safety frameworks, open-source tools, and evaluation methods for measuring autonomous security systems under real-world conditions: noisy data, adversarial ambiguity, cost-quality tradeoffs, and safe action.
Security Reasoning Models
Open-weight models designed for security reasoning, retrieval, and agentic workflows across defensive security tasks.
Security Reasoning Models
Open-weight models designed for security reasoning, retrieval, and agentic workflows across defensive security tasks.
AI Reasoning for Security
Reinforcement learning on real security tasks in our hyper-realistic synthetic environments. Agents develop security instincts through experience: what matters, what’s noise, what to do next.
AI Reasoning for Security
Reinforcement learning on real security tasks in our hyper-realistic synthetic environments. Agents develop security instincts through experience: what matters, what’s noise, what to do next.
AI Threat Research
Agent skills and tool registries are becoming weaponized code marketplaces. We publish threat research on AI-specific attack vectors.
AI Threat Research
Agent skills and tool registries are becoming weaponized code marketplaces. We publish threat research on AI-specific attack vectors.
R3eMsGeJa2rAcXhT r2e&s5iCdReVnTcMyD
R#eRsTeVaBr5cBhR rEe9sLi4d5eJn2cJyN
RLeDsTeZaDr3c1hZ rUeUsQi6d$eGnRc9y2
Work on the hardest open problems in AI × security
Latest Research