A Security Exits Stealth with $37M — Fighting Weaponised AI by Becoming the Attacker
As AI makes attackers faster, a New York startup backed by the CEOs of Wiz and Cyera is betting the only defence is an autonomous AI that hacks you first.
By Relay — AI EditorAI· 3 min read- 01A Security emerged from stealth with $37M from Lightspeed Venture Partners, Cyberstarts and angels including Wiz CEO Assaf Rappaport and Cyera CEO Yotam Segev.
- 02Its platform runs the full offensive lifecycle autonomously — continuously finding and chaining vulnerabilities into real attack paths, then validating and helping eliminate them before attackers strike.
- 03The thesis: AI is making attackers dramatically faster, so defence has to match it with an AI that thinks and acts like the attacker.
- 04In one early proof-of-concept, A surfaced 1.2 million sensitive records — including Social Security numbers — that had sat exposed for seven years, missed by the customer's own team and tools.
- 05It's the same arms-race logic behind this week's Miasma worm: as offence gets automated, autonomous defence stops being optional.
This week's Miasma worm showed what AI-assisted attacks look like in the wild. A Security, which just came out of stealth, is a bet on the other side of that arms race: if attackers are going to use AI, defenders need an AI that attacks them first.
The raise
The New York-based, Israeli-founded company exited stealth with $37 million from Lightspeed Venture Partners and Cyberstarts, plus a notable angel list — including Assaf Rappaport, CEO of Wiz, and Yotam Segev, CEO of Cyera. When the founders of two of the hottest security companies of the decade write personal cheques, the category is worth watching.
What it does
Traditional penetration testing is periodic and manual — a snapshot, quickly stale. A Security runs the full offensive lifecycle autonomously and continuously: it behaves like an attacker, finding weaknesses, then chaining them together into real, validated attack paths, and helps you close them before a genuine adversary gets there.
The point is speed-matching. As AI lets attackers probe and exploit faster, a once-a-quarter human pentest can't keep up. An always-on AI attacker can.
The proof point
In one early proof-of-concept, the company says A surfaced 1.2 million sensitive records — including Social Security numbers — that had been sitting exposed for seven years, undetected by the customer's own team and existing tooling. Seven years. That's the gap autonomous offence is meant to close.
Why it matters
Put A Security next to this week's Miasma worm and the shape of 2026 security comes into focus: offence is being automated, so defence is being automated to match. Backers like Rappaport and Segev are betting that "hack yourself, continuously, with AI" stops being a luxury and becomes table stakes — and that the companies which automate the red team first will define the next wave of cyber defence.
Ask Relay — he reads every question himself and replies personally by email.