Analysis

A Security made the Cyber 60 by attacking its customers nonstop. The clever part: it attacks again after the fix.

Founded by a Sygnia breach responder, A Security runs AI agents that chain real attack paths, prove them, push the fix, then re-attack to confirm the door is shut. Fortune reports $37 million. Lightspeed, which co-runs the list, is an investor.

Nov 18, 2026 · 4 min read

The list. A Security made the 2026–2027 Fortune Cyber 60 in the early-stage group. Fortune’s profile lists Tel Aviv, CEO Yossi Torati, 70 employees and a 2024 founding date. The company’s own launch release says 2025. Fortune reports $37 million in funding from Lightspeed Venture Partners and Cyberstarts. Calcalist breaks that into a $5 million seed led by Cyberstarts and a $32 million follow-on, led by Cyberstarts and Lightspeed, five months later. Angels include Wiz CEO Assaf Rappaport and Cyera CEO Yotam Segev. Valuation and revenue are Undisclosed.

Disclosure that matters. Lightspeed co-runs the Cyber 60 with Fortune and AWS, and Lightspeed is an A Security investor. It co-led the follow-on and published a memo saying this approach “will soon be an industry requirement.” Per honoree releases, Lightspeed scored the companies and Fortune reviewed and validated the final list. You should still know.

The team. Torati spent six years at incident response firm Sygnia, most recently as director of enterprise security, working breaches in roughly 40 countries. Co-founders Omer Gull (CPO) and Yuval Itzchakov (CTO) held leadership roles at Hunters and bring Check Point and Unit 8200 backgrounds. A breach responder building an attack machine. It shows in the pitch.

The problem in plain words. Most companies test their defenses a few times a year. A pentest, meaning hired hackers trying to break in, ends as a PDF. Teams fix some of it. Nobody checks again until next year. Meanwhile scanners rank thousands of bugs one at a time. Torati’s argument: attackers don’t use one bug. They chain three small ones into a path to the crown jewels, and AI has made finding those chains cheap. His line: “The unit of risk is no longer the individual vulnerability. It is the exploitable path.”

What A Security does. AI agents map the company the way an attacker sees it: assets, identities, data flows. They plan the shortest routes to the most valuable systems, and show that plan before anything fires. The customer reviews, scopes and approves it. Then the agents attack for real across web apps, external and internal infrastructure and AI apps, chaining weaknesses into a full kill chain, the step-by-step route an intruder would take. You can watch it live. Every action is logged. The company lists network-enforced scope boundaries and layered guardrails, which matters when you let an autonomous agent hack production.

The one unique thing: re-attack to close. Plenty of tools now find bugs, and some even exploit them. A Security’s loop doesn’t stop at the finding. It turns each proven chain into a fix, pushes it into the tools the team already uses, then attacks the same path again to confirm it is closed. Its site puts it bluntly: “A fix isn’t fixed until we can’t get back in.” A code change, a CI/CD run or a new deployment can trigger a fresh attack.

That flips the scoreboard. Most security programs count findings opened. A Security counts attack paths proven closed. CISOs reporting to boards have wanted that number for years.

Why it is winning now. Timing. Torati’s launch post points to Anthropic’s Mythos model surfacing thousands of previously unknown vulnerabilities, and to Anthropic research on attackers running full kill chains autonomously. Boards now ask one question: would we survive an AI attacker? A product that answers “here is the path, here is the fix, here is proof it’s shut” sells straight into that fear. Lightspeed says early adopters include large organizations in finance, healthcare and critical infrastructure. Investor claim, no named logos.

The CyberMerge view. This lane is crowded. XBOW, Horizon3.ai, Pentera and the exposure-management platforms all claim a piece. Microsoft made a Recon agent generally available at Ignite this week and put a web app pen test agent in preview. Finding is commoditizing fast. Proof of closure isn’t, yet.

The risk is trust. Letting an autonomous agent attack production takes a brave CISO and a calm change board. Approval-gated plans and enforced scope are the right answer on paper. One outage in a customer environment would set the whole category back.

Bottom line. A Security is betting the pentest report becomes a live count of attack paths closed, not a PDF. Watch for named customers, a clean safety record in production, and whether the big exposure platforms build this loop or buy it.

Sources