Analysis

Cogent Security made the Cyber 60 by skipping the easy part. It doesn’t find your bugs. It fixes them.

Ex-Abnormal and Coinbase operators built AI agents that figure out who owns a vulnerability, write the patch and prove it closed. Its own repair model just topped a public benchmark. Fortune reports $53 million.

Nov 19, 2026 · 4 min read

The list. Cogent Security is on the 2026–2027 Fortune Cyber 60 in the early-growth group, its second straight year on the list per Fortune’s profile. Fortune lists San Francisco, CEO Vineet Edupuganti, 72 employees and a 2024 founding date. Fortune reports $53 million in funding from Greylock and Bain Capital Ventures. That matches the company: an $11 million seed from Greylock at its July 2025 launch, then a $42 million Series A in February 2026 led by Bain Capital Ventures, with Greylock and Definition. Valuation and revenue are Undisclosed. Lightspeed, which co-runs the list, is not listed as an investor.

The team. Edupuganti and CTO Geng Sng came from Abnormal Security. Co-founder Thanos Baskous ran infrastructure at Coinbase, including large-scale vulnerability fixing. That last part matters. Most vulnerability startups are built by people who found bugs. This one was built by people who had to fix them.

The problem in plain words. Finding vulnerabilities is cheap now. Scanners spit out thousands. Cogent’s release counts more than 48,000 new CVEs published in 2025. The slow part is everything after the scan: who owns this server, does this bug actually matter here, who needs to do what by when, and did they really do it. Security teams chase owners through stale spreadsheets and Slack threads. Edupuganti told Fortune the hard part isn’t finding. It’s ownership.

What Cogent does. It sits on top of the scanners and tools a company already pays for, plus asset lists like ServiceNow. Its agents gather context on who touches which systems, estimate how likely a bug is to be exploited and how far an attacker could get, then route a specific fix to the right person. In some cases they open the code change themselves. Customers set approval rules. Most start by letting Cogent triage and route, then grant what Edupuganti calls “slices of autonomy,” often in test environments first.

The one unique thing: it built its own repair model. On Nov 3 Cogent released VR-1.1, a model trained to fix vulnerabilities. On LiveCVEBench, a public benchmark built by university and industry researchers that turns fresh CVEs into repair tasks, VR-1.1 produced working fixes for 57.9% of 190 tasks. Claude Code and Codex each scored 45.3%. A fix only counts if security tests show the hole is closed and functional tests show the software still works.

How it works is the interesting bit. The model first writes its own exploit to trigger the bug. It traces related code paths with the same weakness. It makes a small, focused change. Then a separate review, in a fresh context, tries to get around the fix. Finally it writes a regression test so the bug can’t quietly come back. Caveats, in Cogent’s own fine print: Cogent ran the evaluation, and inference budgets differed across the three systems. Treat it as a strong company claim, not an independent result.

Why it is winning now. AI made finding bugs fast this year. Fixing didn’t speed up. That gap is the whole market. Cogent says Fortune 500 customers cut the exposure window for critical vulnerabilities by 97% on average and that it works with dozens of Fortune 1000 companies. Company claims. Bain’s Enrique Salem says teams do five times more with the same people. Investor claim.

The product keeps widening. October brought Attack Path Analysis, which looks for the single change that cuts several attack routes at once. On Nov 10 Cogent bought Specular, a Google-backed startup that maps a company from the outside the way an attacker would. Terms were not disclosed. Specular founder Peyton Smith, five years at CrowdStrike Services, is now Cogent’s head of offensive security.

The CyberMerge view. Tenable, Qualys and Rapid7 own the scanners. Cogent is betting the scanner is the commodity and the fix is the product. That is a direct shot at the exposure-management pitch every incumbent is making. The platforms also want this job. Palo Alto told investors last night it is training models for more than 35 specialized security use cases, and it just agreed to buy a model-training shop to do more.

The risk is the change board. An agent that writes patches touches production code. One bad auto-fix that breaks a payment system and the autonomy slider goes back to zero. Watch for an independent benchmark run, named reference customers, and how many fixes ship with no human edit.

Bottom line. Every vendor sells you a longer list of problems. Cogent is selling a shorter one.

Sources