TikTok users demand Marcus Hutchins open-source his anti-troll AI bot on GitHub
The bot’s job is to be "calm and annoying".
Image by Cybernews. Source: Threads/Marcus Hutchins
Follow us
- Marcus Hutchins built a bot that uses a local AI model to answer suspected social media trolls.
- The bot replies calmly to keep trolls engaged and away from people who may be harmed by harassment.
- Hutchins ran the experiment for three days and said some trolls kept responding without noticing the bot.
- Many users asked for the code, but Hutchins said scaling the bot would break platform rules.
Key Takeaways by nexos.ai, reviewed by Cybernews staff.
A newly launched bot feeds social media troll replies directly into an AI model to generate endless, automated responses. Fans want its creator, famous for halting the WannaCry cyberattack, to make it available for all.
Security researcher Marcus Hutchins, known online as MalwareTech, has created a bot that feeds social media troll replies directly into an AI model to generate endless, automated responses.
Hutchins has been an outspoken critic of AI, arguing that overuse is costing some of his friends their brain cells. But it looks like he's finally found a use for the tech that he can get behind: turning AI against his own trolls.
In a video posted on TikTok and Threads, Hutchins explained that while the standard advice is to ignore internet trolls, this often leaves them free to target vulnerable users.
“Trolls tend to spray and prey,” he said.
The kind of people best equipped to deal with trolls are those who just ignore them, which results in trolls harassing people who are actually affected by their behavior.
So Hutchins decided to create a bot that feeds the trolls their own medicine with automated AI-generated replies.
Every time someone left a mean comment on one of his posts, Hutchins would check their profile.
If it looked like they were serial harassers, he’d add their name to a watchlist so his program would automatically catch any future replies from them and send them straight to the bot.
Bot’s job is to push troll into responding
Hutchins explained that the script would then feed the reply to a local LLM, giving it the following instructions:
You are a calm, confident human being…You are dealing with a troll, and your job is to be as annoying as possible while still keeping them engaged,says prompt to AI written by Marcus Hutchins, to guarantee reply from troll.
Explaining the psychology behind the prompt, Hutchins says, “Trolls are usually deeply insecure. So their job is to drag you down and make you feel bad about yourself.
“So the more calm and confident you present, the more annoyed they get. And the harder they are going to try and troll you.
“The goal of the bot is one that is going to guarantee a reply from the troll.”
Engage harassers in a constant loop
In all, Hutchins ran the experiment for 3 days before posting his “reveal” on the platform Threads last week – a post that went viral, garnering 8,000 likes.
During this time, he noticed different patterns emerging among the trolls that the bot interacted with.
One unsuspecting troll logged on for about 20 minutes a day and engaged with the bot.
After Hutchins published the experiment, the troll saw the post and responded by setting his account to private.
The bot got the best results with people who spent all day online trolling people.
The most "insane" instance, Hutchins said, came on the thread where he openly revealed he was using a troll bot.
He explained that the troll started trolling Hutchins on his reveal post, but then got trolled by the AI bot himself.
After a while, Hutchins stopped the bot manually, curious to see how the troll would react.
“I left a comment that said, ‘You are aware that this is just a bot, right?’ But he just kept going. He didn’t seem to realize it, which goes to show how unintelligent a lot of these trolls are.”
The same troll also tried to dig into Hutchins' colorful background, including a screenshot of a 2020 Wired article detailing his life story.
"He was trying to dig into details about my life – but these are facts that I myself disclosed to the journalist, so they're not things I'm going to be insecure or ashamed about, because I divulged them in the first place."
Post goes viral, people want the code
Responses on Threads and TikTok have been largely positive, with posters hailing him a “Troll slayer” and encouraging him to post the code on GitHub.
“I think people would pay good money for this service," posts Fade transition.
There were also multiple calls for Hutchins to share the script:
“Please tell me ur code is on github. Link it please,” said @Sillyant.
Hutchins said he wishes he could scale up the bot, but doing so would violate the platform’s terms of service.
Hutchins says, "It would be really cool to divert all of the trolls into arguing with bots, so they can't find real victims who aren’t well equipped to deal with trolls.”
Hutchins was hailed as a hero after he was credited with stopping the WannaCry malware attack, which threatened the NHS and other organizations in May 2017.
He was arrested by the FBI a few months later at a Las Vegas airport, after attending the Def Con hacking conference, on charges related to creating and distributing malware.
He has since become an independent security researcher and has recently been a vocal critic of "AI doomers," pushing back on claims about the dangers of frontier AI models – including Anthropic's Mythos – which he argues are overstated and not worth their cost.
