AI safety for builders hackathon

Hi Ankur,

Thanks — I agree that the hackathon should remain broad enough for the participants themselves to reveal promising product and tool directions. I was thinking of one optional real-world challenge alongside that open exploration, not of narrowing the whole hackathon.

One need that comes directly from our own audio and transcription work is:

Source-Bound Audio Challenge — Can an AI-assisted audio derivative travel without being mistaken for its source?

An original recording may pass through transcription, restoration, source separation, human correction, and further analysis. Each step can be useful, but the resulting objects can easily become detached from the exact source or silently presented as if they were the original.

Builders could create a small local tool that:

• preserves the original audio bytes unchanged;
• binds transcripts, corrections, and analysis copies to their exact source or parent;
• records hashes and transformation steps;
• detects a source swap, transcript mismatch, or edited derivative mislabeled as the original;
• produces a simple human-readable provenance card.

A small synthetic or openly licensed audio fragment would be enough. The challenge would not need to determine speaker identity or prove complete audio authenticity. The narrower question is whether provenance remains visible while the audio travels through AI-assisted transformations.

This feels like a real product-shaped need rather than a toy problem, while still leaving plenty of room for builders to approach it creatively.

Does this direction feel suitable for the kind of participants and hackathon you have in mind?

Loek

@Loek Verdonk Thanks for sharing. This is an interesting problem sure - but only sharing this as the hackathon statement will probably greatly limit intended audience for now.

I can consider listing it as one of the problem and attribute to you.

@Ankur Pandey

Wow, that would be amzing Ankur.
Because I never meant it as the main hackathon statement, only as one optional real-world problem alongside the broader set.

Thank you for considering listing it with attribution. I want to make this as easy as possible on your side, so when it becomes useful I will prepare a complete, ready-to-use package: a one-page challenge brief, clear acceptance cases, and a small synthetic or openly licensed audio fixture that participants can build against directly. No extra work needed from you or your team — you decide whether and how it fits.

This problem comes straight from our own daily work, so whatever gets built around it will be genuinely used, not just judged.

Good luck with the hackathon!

Loek

@Loek Verdonk done! Will touch base once hackathon begins to materialize. tx