Founding Member of Technical Staff, Security and IT

About LH2 AI Labs
LH2 AI Labs builds post-training infrastructure for frontier AI models. Our mission is to bring data that has never been used for training (the code, conversations, documents and workflows inside real institutions) into frontier AI pipelines, safely and with the trust of the people who own it. We work across verticals such as coding, computer use, agentic workflows, medical and audio, alongside a network of vetted domain experts. We're a small team in Bengaluru, founded by second-time founders.
About the role
Our business depends on two groups trusting us with things they protect closely. Companies trust us with their internal code and data. Frontier labs trust that what we deliver is clean, has rights cleared, and won't come back to hurt them. Every promise we make to either side is a security claim, and we want each one to be true and provable.
You'll be our first dedicated security engineer, and you'll own security across all of it: company IT (identity, devices, SaaS and the people who use them), our cloud infrastructure, the software we run inside data owners' environments, and the codebases and datasets that move through us. We want someone who builds security systems rather than configures products, and who treats the corporate laptop and the production cluster as parts of one attack surface. You'll write code, design controls, and be the person a customer's CISO wants on the call.
Responsibilities
Company IT
Own identity for everyone at LH2: SSO, MFA, groups and roles, and access that follows the person from joining to leaving.
Manage and secure every device that touches company data, including device management, disk encryption, patching, and checking device health before granting access.
Automate onboarding and offboarding so access is granted on day one and fully revoked within the hour someone leaves.
Keep an inventory of the SaaS tools we use and who can reach what in each.
Secure email and collaboration tools against phishing, spoofing and accidental oversharing.
Give contractors, vetted experts and vendors the access they need, on devices we don't own, without exposing anything else.
Cloud infrastructure
Design our cloud account and network structure: separate environments, segmentation, and no long-lived credentials.
Manage infrastructure as code with security review built in, and keep secrets and encryption keys in managed systems with clear owners.
Centralise logs from cloud, identity, endpoints and applications, and build detection on top of them.
Harden CI/CD, containers and build pipelines, and put security checks where engineers already work.
Across the company
Own LH2's threat model, including insiders, our expert network, compromised customer environments and the buyers we deliver to.
Secure the redaction kit and other software that runs inside data owners' environments, from build and signing through update and removal.
Build chain of custody for every codebase and dataset, from intake to delivery and withdrawal.
Lead SOC 2, DPDP and lab security reviews, with evidence generated by systems rather than spreadsheets.
Own incident response, and sit in security reviews with data owners and labs.
You may be a good fit if you
Have run identity and device security for a company: an identity provider, device management and access reviews, not just one of these.
Have 6+ years in security, infrastructure or IT engineering, with real ownership of security outcomes rather than only advising on them.
Know AWS or GCP IAM, networking, containers and CI/CD well enough to break them and to fix them.
Automate what others do by hand, in Python, Go or a similar language.
Can find vulnerabilities in both code and design, and explain them to the engineers who wrote them.
Can turn a compliance requirement into a control that actually reduces risk.
Treat "how do we know it worked?" as part of the job.
Are comfortable being the person a customer's security team questions.
Strong candidates may also have experience with
Building company IT from scratch at a startup, or being its first security hire.
Zero-trust access: device-aware access, identity-aware proxies, replacing VPNs.
Insider-threat programmes, data loss prevention or privileged-access systems.
Software that runs on customers' machines or inside their networks.
Taking a company through SOC 2, ISO 27001 or DPDP readiness for the first time.
Offensive security or red teaming.
Representative projects
Make offboarding a single action: one change in the identity provider removes access from every SaaS tool, cloud account, repo and device within the hour, and produces a record proving it.
Put every laptop that touches customer data under management, and block access from any device that isn't.
Give vetted experts a browser-based workspace for sensitive data, with no download, no copy-out, and every session logged.
Restructure our cloud so production data sits in its own accounts, reachable only through time-bound, approved access.
Finish our access audit and make "nobody at LH2 gets access they don't need" true, in a form a data owner's auditor would accept.
Threat-model the redaction kit so a data owner can approve it: signed builds, no outbound paths except the ones they allow.
Fingerprint every delivery so that a dataset leaked from any lab buyer can be traced back to its delivery.
Automate SOC 2 evidence collection so the audit is a byproduct of how we operate.
How we're different?
We think the next gains in AI will come less from more of the public web and more from data that has never left the institutions that produced it. Getting that data is mostly a trust problem, and we treat trust as an engineering problem. Data owners keep their raw data. We measure our claims rather than assert them. Nobody at LH2, including us, gets access they don't need.
We work as one team on a small number of hard problems rather than many small ones in parallel. We'd rather ship one thing a frontier lab relies on than many things that look good in a demo. We write things down, argue about them, and change our minds when the evidence says so.

Work arrangement
No