Company

We build offensive security
that engineers trust.

Pentesting hasn't kept pace with how fast software now ships. Manual engagements take weeks; commodity scanners drown teams in false positives. We're building the third option. Fast, autonomous, and honest.

Who we are

Zeroday IQ Cyber Pte. Ltd.

A registered Singapore private company building offensive security software. We operate out of Republic Plaza in the heart of the Raffles Place financial district, and we ship from there to engineering teams worldwide.

We were founded by offensive security and applied ML engineers with a simple thesis. Software ships every day, but pentests still take six weeks. That mismatch is how breaches happen. Zeroday IQ closes the gap with an autonomous AI red team that runs continuously, validates every finding with a real PoC, and never invents a vulnerability the scan didn't actually produce.

All commercial engagements, contracts, and data processing agreements are entered into under Zeroday IQ Cyber Pte. Ltd., governed by Singapore law.

Legal name

Zeroday IQ Cyber Pte. Ltd.

Status

Live Company, ACRA Singapore

Registered office

9 Raffles Place, #29‑05
Republic Plaza
Singapore 048619

What we believe

Three principles, no exceptions.

Every line of code in Zeroday IQ (every scanner module, every prompt, every UI choice) is shaped by these.

1. Zero false positives

If we can't show a real PoC, it doesn't make the report. Active scanners run baseline + control diffs instead of bare fingerprints. The Validator agent is the last gate; it drops anything that doesn't have raw evidence behind it. False positives are how scanners lose trust. We don't ship them.

2. Grounded AI

Models are tools, not oracles. Every analysis call runs against a locked system prompt that forbids invented findings and demands a proof field on every issue. Chat answers are bound exclusively to your stored scan. The LLM literally cannot invent a finding the scan didn't produce.

3. Talk to it, don't operate it

You shouldn't need to learn a tool to run a pentest. Type a URL, ask follow‑ups in plain English, get answers backed by real evidence. The whole product is a conversation. No CLI, no Burp config, no scoping calls before you can see your first finding.

What we built

A pentester's pipeline,
driven from chat.

Zeroday IQ is 41+ independent scanner modules orchestrated by a thread‑pool, plus thirteen specialised AI agents. Seven for external scans (Recon, Enumeration, WAF Detection, Vulnerability, Exploit, Validator, Report) and six more for internal‑network engagements (Network Discovery, Active Directory, Credential, Lateral Movement, PrivEsc, Attack‑Path / Blast‑Radius / MITRE).

Every finding is tagged with CWE, CVSS 3.1, OWASP Top 10, MITRE ATT&CK, and seven compliance frameworks: SOC 2, ISO 27001, HIPAA, GDPR, PCI DSS, NIST CSF, OWASP. The PDF ships the moment the scan completes.

41+

Scanner modules. DNS to SSTI, every layer of the stack.

13

AI agents, each with its own toolset and reasoning trace.

100%

Recall on the OWASP Juice Shop benchmark.

0

False positives. By design, not by aspiration.

Benchmark scorecard
2026 Q1 · 5 targets
passing
Z
OWASP Juice Shop · recall 100% · 0 FP DVWA · recall 92% · 0 FP WebGoat · recall 88% · 0 FP bWAPP · recall 84% · 0 FP Vulnerable SaaS · recall 79% · 0 FP Average: 88.6% recall · 0 FP across 5 targets
Honest about limits

We tell you what we
can't yet find.

Our public benchmark scorecard tracks exactly what we recall and what we miss. Currently CSRF detection in some flows, file‑upload validation edge cases, and Java deserialisation gadget chains. We're shipping fixes monthly and our changelog is public.

Most scanners hide their gaps. We publish ours. If we miss a known finding on your benchmark, that's a bug. File it and we'll fix it.

Company

Small team. Big mandate.

We're a focused team of offensive‑security engineers, ML practitioners, and product builders. We ship every week.

Offensive security

Years in red‑team consulting and bug bounty. We know what a real pentest report looks like, and what makes one useless.

Applied ML

Production LLM systems with structured outputs, prompt caching, and groundedness guardrails. We treat models like dependencies, not magic.

Distributed systems

Parallel scan orchestration, MongoDB‑backed state, agent fleet management. Built for engagements that touch hundreds of subdomains in parallel.

Want to see how we work?

Sign in and run a scan on your own domain in the next ten minutes. No card, no scoping call.