Company · About
We build the evidence layer AI approval was missing.
AetherLab is an AI risk company in San Diego. We red-team AI systems, contain their failures in production, and turn the results into evidence that risk teams, underwriters, and procurement committees can act on.
Started in adversarial research. Forged in production.
AetherLab's founding research question predates the current AI wave: how do you know when a machine-learning system will hold up against inputs designed to break it? Our founder's doctoral work on the adversarial robustness of classifiers was peer-reviewed at IEEE ICMLA in 2019, before adversarial AI was a commercial category.
The commercial insight came later, from the other side of the table. Working on applied AI at BCG GAMMA and risk systems at Gemini made one thing obvious: the hard part of enterprise AI was never the demo. It was the approval: the moment a risk team, a bank, or a regulator has to decide whether the system is safe enough to stand behind, with no standard evidence to decide on.
So that's what we build. AdversarialScan finds how a system breaks. Guardrails hold the line in production. The Evidence Pack turns both into a record an institution can approve. AetherLab processes 15 million guardrail queries per month.
What we believe
- Evidence over assurances. Every claim we make about a system is reproducible; every number we publish is measured.
- Severity over pass/fail. Real risk decisions need ranked findings tied to consequences, not a green checkmark.
- Depth over theater. We publish our research where we can, keep our attack methodology private on purpose, and skip the logo walls.
Built by
- Published adversarial-ML research (IEEE, 2019)
- PhDs in physics, statistics, and computer science
- Alumni of BCG, McKinsey, and Gemini
- Production AI risk systems in payments and fintech
- Advised by a former CISO of Twitter, CrowdStrike, and F5 and a former Lloyd's Syndicate 1200 active underwriter
Meet our founder.
Physics PhD (UC San Diego, 2019) whose published research covered the adversarial robustness of machine-learning classifiers, peer-reviewed at IEEE ICMLA in 2019 before adversarial AI was a commercial category. Worked on applied AI at BCG GAMMA and on AI-driven risk systems at Gemini before founding AetherLab.
- ·Published adversarial-ML research (IEEE ICMLA 2019)
- ·PhD in physics, UC San Diego
- ·BCG GAMMA · Gemini
The work is on the record.
Our published research is in adversarial robustness and model evaluation: the defensive science. The attack methodology inside AdversarialScan is proprietary and deliberately unpublished; the credentials behind it are not.
A. Georges et al. · IEEE ICMLA 2019 · DOI ↗
Peer-reviewed work on robust classification: adversarial-ML research published before adversarial AI was a commercial category.
A. Georges · BCG GAMMA (Medium)
An accessible account of the shape-of-data methods behind robust classification, including what adversarial examples reveal about model fragility.
A. Georges · UC San Diego, 2019
The research foundation: mathematical structure as a lens on model behavior and robustness.
S. Kumar, N. Beckage, et al. · NeurIPS 2024 workshop (Red Teaming GenAI)
Team research on adversarial prompting and automated methods for evaluating bias in language models.
J. Qiu et al. · Neural Computing & Applications, 2023
Team research on measuring the internal representations neural networks actually learn.
Work with the team.
Approving AI, building it, or writing about the space: we respond within one business day.
Ask about the Evidence Pack
Leave your email and we'll walk you through what an Evidence Pack contains for your use case: severity-scored findings, business-impact mapping, and the approval record.
