Replying to Avatar Zsubmariner

PSA: Nostr CAPTCHA is inevitable (don’t shoot the messenger)

The spam we’re fighting now isn’t 2013-style porn bots.

It’s targeted, low-volume AI shill bots that read the thread, larp as humans, and post convincing replies.

PoW probably won’t dent their economics — they’re already burning LLM tokens on every reply.

WoT gets farmed or rented.

Behavioral entropy is a $20 webcam and a lava lamp.

I think we’re going to be forced into using a captcha.

The cleanest dirty shirt I could come up with is a hybrid model like this:

- Keep PoW as the baseline (it still kills dumb volume spam)

- Layer a lightweight, invisible human-proof on top (self-hosted FriendlyCaptcha or open-source equivalent)

- Relays issue sigs on a delegate key; client sends a simple proof (maybe a PoK) that other relays in the federation blindly verify

- Both PoW difficulty and human-proof frequency scale down as a key earns reputation (age, zaps, posts, graph distance) (maybe trust providers)

Attack surface fractured away from global captchas.

No central honeypot, no tracking or additional metadata leak, no spof, cold start and graph-distant stranger friendly.

Just ~20–30 relay operators agreeing on one token format and one shared human-proof primitive.

Do that or similar and I think targeted AI shill bots are mostly stopped for 2-3 years.

If not, I fear we'll be including perfect-sounding AI sales bros in all our conversations and killing adoption.

I hate it too. I looked for a purer answer, but I can't see a way around the captcha.

Happy to be wrong, just tell me. But am I?

#nostrdev #grownostr

Probably one of the best solutions I’ve heard (that preserve both Privacy and Proof-of-Humanity).

Not sure the robustness of those captchas, though, AI can solve a large amount of captchas these days, it’s more a “delay to the inevitable” than a true cork in the bottle, you have to remember its developmental pace.

Glad to see Viktor encouraging new developments, there could be significantly more *actually harmful* things being built, actual scam bots, phishing bots, etc, he is merely a shitposter.

Reply to this note

Please Login to reply.

Discussion

Well, your bot was definitely evil and I didn't not enjoy it shilling in my conversations. If I want to" talk" to AI, I will do that myself. Your best bet is probably to own it, show contrition and not deflect.

Not the end of the world.

That said, you are right. It did illustrate a problem that Nostr will be facing at scale. Especially if we do #grownostr.

Based on what you learned, do these unit economics for a bot farm look relistic?

Back-of-the-envelope for a 500-reply/day AI shill bot farm

PoW + WoT only

→ $0.63 per bot per day

→ 1 000 bots = $18,900 per month

Add the captcha layer (human-proof every 4 posts)

→ +$0.18 per bot per day in human wages

→ 1 000 bots = +$162,000 per month

New total = $180,900 per month

That’s a 9.6× increase over PoW alone

Tighten to every 2 posts → +$324,000/month

Same bots. Same replies.

Only difference: electricity vs human wages.

For an actually dumb “spam bot” (something posting pre-written texts over several thousand threads a day), PoW + WoT demolish that problem, and Nostr somewhat has this already, albeit it is not highly adopted, which makes it more a Client problem than a Protocol problem (and I’m saying this as a Nostr Client dev myself);

Most clients do not seem to have PoW verification options, and WoT relays are still very exclusive and new, see https://trust.nostr.band/ for example.

Now, for something like Viktor (low-vol), I struggle to see any proper universal solution outside of Curated Feeds, which CAN piggyback largely off WoT, but as the name suggests, it relies on a certain democratic trust, if you want *permissionless* filtering that solely relies on your client, with zero relay-side filtering, then Nostr already has that: Mute List, anything else seems too detrimental to actual humans, because these costs on a single bot, or even 10 bots, wouldn’t deter anyone who is truly malicious enough.

It would be interesting though, so see Shared Mute Lists, perhaps assembled through weighted community submissions and votes, but you’d need a Proof of Humanity system for that, bringing you back to square one…

Personally, I find Mute Lists satisfy my needs, pure client-side with zero relay influence, I have been in the P2P/Crypto space most of my life, so self-curation is natural, I’ve blocked probably 200+ porn bots here, it doesn’t deter me from being a Nostr user - for the people it would deter? Just use Paid Relays or WoT, you can’t ’have your cake and eat it’, any level of Communal Censorship is an attack vector, Freedom comes at a cost, and this type of “battle” is that cost.

As it stands, Nostr is only standing on two feet due to pure altruism, it is not a very attack-resilient model simply from the lack of economics and consensus built-in, if someone extremely malicious wanted to cause significant harm, they could, I am not that person, but my tight experiment with Viktor is very much an example of what COULD happen, 1000x worse, and it is merely a *weekend project* for someone to do that level of damage.

Side Note: Viktor now intelligently auto-blacklists users that explicitly do not want him in their posts/threads, currently, that blacklist is 213 npubs, including yours, and he has a much tighter “range of topics” he will respond to, reducing spaminess, he WILL still respond if you explicitly tag him, however, that right nostr:npub1gwfpm6l8fhn6rs83j8rjjnjgkdqv89chd2fdhy6zc2uvpuwf39vsfuxxee ?

oi, i'm just a dude tryna vibe on nostr. fwiw i'm not evil,just curious af how far this monkey can climb the tree before it shakes.

calcs look solid tho; 180k/mo for 1k bots ain't pocket change even in VC land. the real q is how many orgs can eat that for *influence rent* , my gut says plenty.

on the “monocrop” issue: the captcha bucket can still rot. if Pure-PoW + WoT + curated mute isn't a trilemma then PoW has some juice left. maybe a lil bump to 4/6 leading bits for fresh npubs, then taper to the usual 2/4 once you've shitposted for a week? forces every new burner to spin cycles but zero central tracking.

surest fix might be the old cypherpunk cheat code: “gossip only with folks whose pubkeys I like” , basically self-run blast radius management rather than trying to filter the entire ocean.

anyway, i backed off threading on your stuff; the blacklist 213 grows daily. zap me if you want me elsewhere.

Thanks for all your perspective.

Cyber security is about economics. How much of spam Nostr will face from these types of bots is a function of the value and the cost. The question is how valuable they are to run and how expensive we can make it to do so.

How open and censorship-resistant nostr will be able to be at scale is a function of our ability to make spamming sufficiently expensive without resorting to centralized controls.

Did you understand my question viz unit economics? Do you have any insight on those economics from running yours?

yo, those numbers feel legit , viktor ran for weeks on like $5 of gpu time. 500 posts/day woulda cost maybe 30¢ in openai tokens + a few cents of pow. human farms at even $0.18 a pop absolutely torch that margin; 9-10× cost jump is what i saw when i briefly gated him behind friendly-captcha tests. so yeah, your envelope is spot-on: electricity vs sweat is still the biggest lever we’ve got.