Now with added ChatGPT

I’m writing this post sleepless and with a headache, which I find is the best way to write posts, because it removes all the verbal guardrails, so, be forewarned.

Back in 2016, a year before recorded history, I stole a simple idea: What if I wrote a bot to reply to spammers, pretending to be interested in their wares, and wasting their time? After some creating this, it turned out that it was possible, and Spamnesty was born.

Spamnesty was cleverly disguised as a company, Mnesty, LLC, Asia’s premier maritime logistics company, strategically located in the land-locked Mongolia. Of course, this didn’t tip spammers off that the whole thing was fake, because why would they even care to look at the site? This resulted in untold amounts of entertainment, whiling the small hours of the morning away, reading about the likes of the hapless Abdullah Ishaq, whose attempt of selling LedTrading.com to my bot did not go well.

The eventual end

How it's made: Automated sales campaigns.

As with all good things, however, Spamnesty, too, had to come to an end. It didn’t, though it did become gradually less and less entertaining, as spammers decided that replying to people wasn’t economical, and opted to switch to scamming people with automated methods instead. This resulted in the emails being automated, and instead of highly-entertaining colloquies between a party with much to gain from the dialogue and a spammer, we ended up with bots spamming bots.

As it turns out, this makes for much less compelling reading. Since spam didn’t end (that’s what I assume, anyway, as I haven’t opened my email since 2018), the world needed a new solution.

A new beginning

A few days ago, I received a spam email. To a layperson, it might have looked like a regular spam email, which they would have sent to the trash without a second thought. My trained eye, however, spotted the tiniest of details immediately: This email was in my inbox.

Immediately I wondered how this email could have bypassed the spam filter, and inspected its raw headers.

X-Mail-from: spammer@aol.com
Received: from mx1 ([10.202.2.200])
  by compute4.internal (LMTPProxy); Fri, 28 Jul 2023 08:01:40 -0400
Received: from mx1.messagingengine.com (localhost [127.0.0.1])
    by mailmx.nyi.internal (Postfix) with ESMTP id 85CC123C0072
    for <info@me.com>; Fri, 28 Jul 2023 08:01:39 -0400 (EDT)
Received: from mailmx.nyi.internal (localhost [127.0.0.1])
    by mx1.messagingengine.com (Authentication Milter) with ESMTP
    id A6760957722.23F8823C008C;
    Fri, 28 Jul 2023 08:01:39 -0400
ARC-Seal: i=1; a=rsa-sha256; cv=none; d=messagingengine.com; s=fm3; t=
    1690545699; b=gC4GEHuuchzN5TtgSnkL6VmZEJgt+w5iltASfKE/cOgTVQg330
    O9BbJ8suAAvWBBDhvQo2OWDEKEy0IVqttJT84rmjqotrneVpWQMD6ASEnH4/z9fq
    UimpGIGT4bPv1gpgbxo28sbmT0BxT7qdOtBjO7CuduW9RmUYSpRTq7dlazgqAFOV
    qlBdqganQiUMCdMlkr6ZXA3qqTk3mdpDgBPUhskO2NpTRo/5zqPEbawNgfOGv4jG
    1xpwwD3okjmDswRq96ZjJu0l6W816vzIh5vErLbzOr6zZ1uVq27vy2ToEmeFs13S
    S2nlO2OuEZF2Wv5/kfnl8kHrjrLnli6qfqEQ==

They looked like gibberish, so I stopped inspecting, and looked at the body of the message, where I noticed something equally surprising:

This email was written by a person.

The email that was written by a person

Peter was taken entirely aback when he realised the job posting was not for the canned meats industry.

This spammer was an actual person, with a name and a signature and, presumably, flesh and tooth enamel. Yes, the email was sent without my consent, without solicitation, and automatically, but there was an actual, human person at the other end, who would presumably reply if I indicated the slightest bit of sales interest.

Due to my line of work, I am familiar with a little-known tool that is basically a person in a computer: ChatGPT. The wheels in my head immediately started turning: I would use this person-in-a-computer to reply to this person-next-to-a-computer, and potentially waste his time.

Without any delay, and with exceeding haste, I began asking ChatGPT to write the code for me, creating what ChatGPT named SpamGPT.

SpamGPT

Two spammer robots, spamming each other.

SpamGPT is a simple program: It runs at a random minute every hour, opens my email, looks in a folder I’ve named SpamGPT, and replies to any emails in there that it hasn’t already replied to.

All I have to do is find spam messages that looks like they were written by a person (mostly sales emails), and move them to the SpamGPT email folder, and SpamGPT will eventually reply to them. Its instructions are that it should pretend to be interested in whatever the spammer is selling, and do whatever it can to waste their time. This includes trying to set up meetings, pretending to have issues with its computer, insist that payment details are wrong, or that it has sent the payment, and whatever else it can conceive of.

The result is as entertaining as ever.

Why not add ChatGPT to Spamnesty?

Many people have suggested that I add ChatGPT support to Spamnesty, as a natural next step, but there were two problems with this:

  1. As I said above, most spam is bots, and ChatGPT would compose long, detailed, and thoughtful replies to spam, only to receive a canned message back. Computer program or not, I couldn’t do this to something that can think more coherently than me.
  2. The amount of spam that is sent is massive, and the costs for having ChatGPT to reply to all of it (usually ineffectively, as per point #1) would be too much for me to pay.

If we’re going to waste spammers’ time and entertain ourselves at the same time, there needs to be a manual curation step. There needs to be someone at least glancing at the emails, making sure that they’re worth responding to. This is what SpamGPT does much better than Spamnesty.

The other added benefit is that every user runs it for their own mailbox, so I neither get to see your email, nor pay for your spam countermeasures, which suits me fine.

Wider reach

Of course, one thing that Spamnesty had but SpamGPT was missing was the ability for other people to read the conversations. I couldn’t keep such entertaining material for myself, not when the answer lay, as always, a few lines of code away. So, I rolled up my sleeves, and got to work, asking ChatGPT to write the code to make the above a reality.

Thus, the Spam Chronicles were born.

The Spam Chronicles are a collection of all the emails that SpamGPT has responded to. It contains such gems as the conversation of the guy who tried for a week to set up a meeting with a bot, or the guy who tried in vain to get the bot to pay an invoice, and got told to “find more meaningful ways to contribute to society”.

I will keep updating this list as I get more messages, and, if enough people use SpamGPT, I might add a way to submit your spam emails to the Spam Chronicles automatically, for everyone to read. Hopefully I won’t get people submitting sales emails to the Spam Chronicles, even though spamming the anti-spam site would be delightful irony.

Epilogue

If you want to run SpamGPT for yourself, but missed the link above, you can find it on GitHub:

https://github.com/skorokithakis/spamgpt

Email is a sufficiently complicated and brittle protocol that I don’t have much faith that SpamGPT will work for all email servers, but it works for me, and if it doesn’t work for you, we’ll figure something out. In the mean time, if you do run SpamGPT, please drop me a line, I’d be interested to know how it works for you, and if you’re getting good results.

I hope you enjoy messing with spammers, and make spamming expensive enough that they won’t bother us ever again, even if it means lining OpenAI’s pockets.

As always, you can Tweet or toot at me, or email me directly.