Cogent Security Gets $42M to Boost AI Vulnerability Defense

An early-stage startup led by a former Abnormal AI division head raised $42 million to better ingest and reconcile massive volumes of enterprise security data.

See Also: AI or Data Governance? Gartner Says You Need Both

The Bain Capital Ventures-led Series A funding will enable San Francisco-based Cogent Security to scale its architecture to handle increasingly complex enterprise environments, reconcile conflicting data sources and improve artificial intelligence reliability in security use cases, said CEO Vineet Edupuganti. The money will address the data and context layer, the AI agent development platform, and the modeling and research layer.

“We spent the last year really building conviction in the space that we’re working in, setting the right foundation in terms of our team, product, early customers,” Edupuganti told Information Security Media Group. “With this round, we wanted to pour our fuel on the fire and double down on the things that are working, and we have plans to scale very aggressively.”

Cogent, founded in 2024, employs 37 people and has raised $53 million, having emerged from stealth in July 2025 with $11 million in seed funding from Greylock Partners. The company has been led since inception by Edupuganti, who spent four years at Abnormal, culminating in an 11-month role leading the company’s 30-peson multi-product platform division (see: Fighting AI Threats With Behavior-Based Awareness Training).

Why Maintaining a Fleet of AI Agents Is Such a Heavy Lift

Edupuganti said the capital is earmarked for significantly increasing headcount. The money will help Cogent to bring its defensive AI capabilities to more enterprises facing increasingly AI-enabled threats, Edupuganti said.

“We’re going to triple the team over the course of this year, invest very heavily into R&D, and at the same time, scale our go-to-market team so we can expand our impact to more customers,” Edupuganti said.

Enterprise security environments are extraordinarily fragmented and data-heavy, and Edupuganti said companies rely on EDR platforms, vulnerability scanners and CMDB systems that generate overlapping, and sometimes conflicting, information. Determining the authoritative source of truth becomes very complex as the number of integrations grows from two or three to a dozen or more, Edupuganti said.

“What we’ve proven is that we can take some core datasets and organize those datasets and help to drive action,” Edupuganti said. “One area that is going to be critical as we mature is scaling to the biggest environments. One thing that’s really critical is architecturally setting ourselves up to achieve high performance at scale. Because generally, performance degrades at scale.”

Nearly every major function at Cogent – from integration management to vulnerability prioritization and remediation planning – is powered by agents, but Edupuganti said building, testing, evaluating and maintaining dozens of agents requires significant engineering effort. Cogent is building a development process that reduces both the upfront cost of spinning up new agents and the variable cost of maintaining them.

“We have AI agents that power pretty much every component of the platform,” Edupuganti said. “That’s things like our integrations, platform reporting, use cases, gathering asset context, prioritizing vulnerabilities, building remediation plans. Basically everything is meant to be an agent. That’s also why we’re able to solve some of the challenges we’re able to solve for that couldn’t be tackled previously.”

Why Foundational Models Fall Short in Specialized Domains

Foundational models excel in code generation and general text transformation but fall short in highly specialized domains like vulnerability management since these workflows often involve contextual nuances and environmental dependencies not well represented in public training data. Achieving the necessary reliability requires post-training refinement, domain-specific data curation and research.

“In terms of getting into the nitty gritty details of what will happen when you try to patch some system, it can be tough,” Edupuganti said. “Because for some of these patterns of context and environmental details, it’s not like there’s some knowledge base that necessarily pre-exists.”

Autonomous remediation will help organizations reduce risk without relying entirely on human-driven processes, cutting down attack paths before adversaries can exploit them, Edupuganti said. Cogent’s strategy is to first deliver operational uplift through automation of labor-intensive triage and analysis tasks, then to gradually guide customers toward greater autonomy.

“If you can know, ‘Here’s the risk performance of an organization and here’s a safe way of mitigating the risk,’ that’s the ultimate unlock for auto remediation,” Edupuganti said. “Can you actually get to a world in which you don’t have to rely on humans and you can cut down attack paths before attackers can break in? That is the holy grail that we’re pioneering towards.”

Customer count is a leading indicator of market validation and expanded impact, but Edupuganti said growth must be matched by measurable product performance. Key product metrics include remediation accuracy, reduction in mean time to remediate and operational efficiency gains for security teams. He said Cogent customers should see significant MTTR reductions and reduced manual workload.

“What is the accuracy of our remediation recommendations? How much are we impacting MTTR for customers?” Edupuganti said. “The mean time to remediate that should be going down and to the right by significant margins as you use Cogent. When you start using Cogent, your security team, your role management team should get way more efficient.”