Fabraix
New accelerator entry · 2026-07-13 · the record
In their words
Fabraix builds state-of-the-art AI red-teaming agents that continuously detect security vulnerabilities in customer-facing AI. Our product, Nyx, has already found vulnerabilities in agents at dozens of Fortune 500 companies. On AgentHarm, the leading benchmark for offensive AI security, Nyx achieved a 78% attack success rate, compared with 67% for GPT-5.6 Sol. AI agents can change more often than teams can test them. A new model, prompt, tool, permission, or data source can change what an agent does, even when the application code stays the same. And AI is also increasing how much software companies produce and how often it changes. We built Nyx to automate the work required to red-team AI agents. It is multi modal by design and interacts with the target indirectly through controlled replicas of SaaS products and websites that we maintain, placing malicious payloads in webpages, documents, files, messages, and tool outputs to test how an agent behaves under different conditions. Nyx often finds its first vulnerability within minutes or hours rather than days or weeks. Because the work is fully automated, companies can repeat the test with every change and at a much lower cost than what a 7-figure red-teaming (and point-in-time) engagement would require. We're also the team behind ACE (Adversarial Cost to Exploit), a benchmark that measures AI security in terms of how much it costs attackers to break an AI system; providing a game-theoretic framework to understand how motivated a rational attacker would be in exploiting the system.
What the record says
Co-founder @ Fabraix, building a frontier hacker agent that finds failure modes and security exploits in AI systems before real users do. Previously the first data scientist at Sequoia-backed Two, where he built AI systems for adversarial environments from scratch and scaled them to process $1B+ in annual B2B transactions, averting $50M in losses for multiple Fortune 150 companies. A published researcher with multiple papers in Q1 journals and a UCL PhD drop-out.
Founders, titles and bios as the company published them. Photographs are our own copies of the ones on that record, so they do not break when the original link expires.
Against the 209 companies on file from Summer 2026. Median team 2, largest 30.
The nearest companies on file from the same batch, side by side with this one.
| Company | Batch | Sectors | Listed |
|---|---|---|---|
| Frontier Computing | Summer 2026 | Artificial Intelligence, Hard Tech, Hardware | 2026-08-31 |
| Ethos Space Resources | Summer 2026 | Hard Tech, Exascale Computing, Energy | 2026-08-23 |
| Spectre Intelligence | Summer 2026 | Biotech, Investing, Trading | 2026-08-22 |
| Edgerun | Summer 2026 | Robotics, Defense | 2026-08-21 |
| Allia Health | Summer 2026 | Artificial Intelligence, Healthcare, Mental Health | 2026-08-19 |
| Fabraix (this page) | Summer 2026 | Artificial Intelligence, Reinforcement Learning, Cybersecurity | 2026-07-13 |
Click a column heading to sort. Comparison is against records held here, not against every company that exists.
The full profile
The rest of this profile is open on any paid or trial plan
It costs no credits. It is a plan check and nothing else.
- Source Check — verify Ahmed Aly's other claims against the registers that hold them.
- INVEST — compare this company against the peers above, on the numbers.
- RECRUIT — see who is already on the team.
- Press Requests — connect with Ahmed Aly for a story.
You land back on this company when you are in.
Not ready to upgrade? Keep this one in view instead.
Watchlist FabraixYou get an alert in MentionFox the next time this company shows up.
Is this you?
If you're behind Fabraix: acceptance into a YC batch is one of the few weeks a young company gets looked up by people who matter — investors, other founders, press. FoxPlug turns the commits, deploys, and releases you're already making into that story automatically, so anyone who looks finds something current. FoxChat puts a guide on your own site so the people YC's own page just sent you don't read one line and leave.
Sourced from the company's Y Combinator directory entry. Figures are read directly from the source record; names have not been matched against other people.