Project ideas from Hacker News discussions.

Meta Ran Ads That Contained AI-Generated Child Sexual Abuse Imagery

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

Four dominant themes in the discussion

Theme	Supporting quote
1️⃣ Scale makes moderation effectively impossible	“it would require millions on millions of employees, spread across ~every country.” – john_strinlai
2️⃣ Cost is used as a pretext; profit drives inaction	“the cost is just a public‑facing excuse.” – lstodd
3️⃣ Executives should be held personally accountable	“Make that the policy, and I'd bet my life savings that Facebook, miraculously, will figure out how to detect this shit with bulletproof accuracy.” – ToucanLoucan
4️⃣ Break up or heavily regulate the platforms	“Just shut the company down, throw the executives in a deep dark hole and forget they exist.” – throwatdem12311

🚀 Project Ideas

MetaAdGuard

Summary

Provides a real‑time AI filter that blocks malicious ads before they reach end users on Facebook/Instagram.
Reduces reliance on Meta’s imperfect automated moderation, directly addressing CSAM and scam ad leakage.

Details

Key	Value
Target Audience	Advertisers, platform moderators, end‑users
Core Feature	Browser extension + API that inspects ad creatives and landing pages for prohibited content
Tech Stack	TensorFlow.js, FastAPI, Cloudflare Workers
Difficulty	Medium
Monetization	Revenue-ready: Tiered API usage fees based on impressions per month

Notes

Echoes HN calls for “shut the pipeline down” but offers a technical alternative rather than platform bans.
Could be adopted by any ad network, creating a universal safety layer praised by regulators and users.

AdVerifiy

Summary

A SaaS platform that enforces advertiser KYC and risk‑scores ad creative for illegal or harmful content prior to publishing on Meta.
Tackles the root cause of harmful ads by imposing a verification gate.

Details

Key	Value
Target Audience	Advertisers, agencies, Meta compliance teams
Core Feature	Automated policy compliance scoring + manual review queue
Tech Stack	Python/Flask, AWS Rekognition, PostgreSQL
Difficulty	Medium
Monetization	Revenue-ready: $0.01 per ad processed + optional premium compliance reports

Notes

Aligns with john_strinlai’s comment that “paid ad campaigns alone is more tractable.”
Would satisfy demands for “requiring approval on individual posts or accounts” from HN discussions.

CSAIDetective

Summary

Decentralized crowdsourced detection tool that identifies CSAM‑like material in ad media using community‑run AI models.
Issues verifiable audit reports to platforms and law‑enforcement.

Details

Key	Value
Target Audience	NGOs, platform auditors, legal investigators
Core Feature	Public API for submitting media hashes, returns risk score + evidence link
Tech Stack	Node.js, PeerJS, IPFS, OpenCV AI models
Difficulty	High
Monetization	Hobby

Notes

Directly responds to quotes like “we would love to see Meta held responsible” from the thread.
Offers practical utility for proving platform negligence without relying on internal moderation.

ModMeta Service

Summary

Scalable “human‑in‑the‑loop” moderation API that reviews flagged ad images/videos for Meta, Instagram, and Facebook.
Combines AI pre‑filtering with vetted annotators to handle the massive volume discussed.

Details

Key	Value
Target Audience	Meta product teams, third‑party ad platforms
Core Feature	Batch processing of ad submissions with turnaround < 24 hrs
Tech Stack	Django, Celery, Azure Cognitive Services, React admin UI
Difficulty	Medium
Monetization	Revenue-ready: $0.05 per reviewed media unit

Notes

Mirrors suggestions that “human moderators” are feasible if costs are passed to advertisers.
Appeals to HN’s frustration about “cost of hiring human moderators” being an excuse, providing a concrete solution.