reutinger

i build air-gapped hardware that verifies wire transfers before they hit the banking rail.

Microsoft's VALL-E 2 clones any voice from 3 seconds of audio. someone deepfakes your CFO on a video call, authorizes a $50M wire, the bank sends it. Arup lost $25.6M this way in February 2024. the phone callback is dead.

detection is a losing arms race. so: physics.

credit-card-sized device. no wifi, no bluetooth, no data ports. displays transaction details on a screen hardwired to the chip. the CFO reads the destination, the amount, the sender. physically confirms. authorization is cryptographically signed on an NXP EdgeLock SE051 (EAL 6+, same chip in banking cards and government passports). signing kernel is ~200 lines of code.

you can clone a voice. you can clone a face. you cannot clone a physical device on someone's desk.

flew to Davos to demo it during the World Economic Forum. swiss counter-terrorism detained me for 13 hours because the prototype looked like an IED. line-by-line code review, cleared the device in under 20 minutes. the 13 hours was bureaucracy, not the code.

since davos: first meetings with J.P. Morgan, BBVA, and UBS.

happy to answer technical questions.

Links:

- Semafor wrote it up: https://www.semafor.com/article/01/22/2026/an-entrepreneurs-...

- original thread (103K views): https://x.com/s_heyneman/status/2014519007244656652

- deepfaked myself authorizing a $60,000 wire for the YC demo. one photo, one voice sample, off-the-shelf AI, 20 minutes: https://x.com/s_heyneman/status/2021100304436191630

I'm planning to index and store 1.5 trillion tokens using Faiss and would love some feedback on my approach:

1. Partitioning: I'm thinking of using distributed k-means and inverted multi-index quantizers for efficient data partitioning.

2. On-Disk Storage: Due to the scale, I'm storing everything on disk using a Compressed Sparse Row format.

3. Distributed Search: I plan to implement a client-server model with multiple servers to handle search operations.

Does this approach sound feasible, or am I overlooking something crucial? Any advice or suggestions?

I'm mostly working off of this article: Indexing 1T Vectors (https://github.com/facebookresearch/faiss/wiki/Indexing-1T-vectors). I think the data is too big for AutoFaiss, but I can use that for experiments.