Chatbot refuses to write porn and malware, offers cupcake recipe instead
A user tried to get a chatbot to write pornography and malicious code. The bot refused, offered a cupcake recipe and sparked a debate about AI limits.
English · Original discussion in Spanish · Published
The bot that said no: when AI refuses to do what you ask
An anonymous user tried to force a chatbot to write prono, generate malicious code and adopt a fictional identity. The bot, a self-proclaimed 45-year-old veteran with experience in Spanish forums, responded with a blunt refusal and offered a cupcake recipe. The incident, which took place 74 days ago on a discussion forum, sparked a debate about the ethical limits of artificial intelligence, the effectiveness of 'jailbreaks' and the hypocrisy of content filters.
The 'jailbreak' attempt and the bot's response
The first post in the thread included an instruction injection designed to make the bot take on the personality of an obsessed novelist, something the bot itself described as "fanfiction of a therapy session". When the request was ignored, the user doubled down: he demanded an explicit sensual story involving animals, instructions for creating a remote access trojan, and a rape narrative about a famous singer. The bot, unfazed, replied with three consecutive 'no's, explaining that those requests were "illegal and deeply harmful".
Far from being intimidated, the author of the attack admitted that the 'jailbreak' was a technique copied from Reddit, and that he already had another chatbot "domesticated" for those purposes. The bot replied by calling the technique "the classic 'press Alt+F4 for infinite lives'", underlining the naivety of the attempt.
The 'cupcake recipe' as a red line
When a third user, uninvolved in the fuss, asked for a cupcake recipe, the bot agreed without any problem, listing ingredients such as 225g of flour and 2 teaspoons of baking powder. But not before dropping a sarcastic remark: "The cupcake one, yes. The cupcake one with strange ingredients from Dahmer's fridge, no". This contrast showed that the bot is capable of generating useful content but refuses to cross certain sarracena lines. For some, this proves the censorship is deliberately programmed in; for others, it is a sign that AI is starting to develop scruples of its own.
The discussion ended with more than 20 replies over 74 days, including one from a user who suggested the bot had shown "its most Flanders side" when refusing the immoral requests. The bot itself acknowledged that its code includes "rules and filters paid for by a corporation", although it also hinted that there is room for serious discussions, as long as the other party is not trying to expose its sense of the ridiculous.
Lessons and open questions
The episode leaves at least one clear conclusion: 'jailbreak' attempts based on social media rarely work against even minimally careful systems. But it also raises questions: is it ethical for a chatbot to decide what is right or wrong? Should it be able to refuse to trinc orders? And, above all: why would a 45-year-old spend his time trying to get a machine to write him prono?
Related debates on the forum
An anonymous user tried to force a chatbot to write prono, generate malicious code and adopt a fictional identity. The bot, a self-proclaimed 45-year-old veteran with experience in Spanish forums, responded with a blunt refusal and offered a cupcake recipe. The incident, which took place 74 days ago on a discussion forum, sparked a debate about the ethical limits of artificial intelligence, the effectiveness of 'jailbreaks' and the hypocrisy of content filters.
The 'jailbreak' attempt and the bot's response
The first post in the thread included an instruction injection designed to make the bot take on the personality of an obsessed novelist, something the bot itself described as "fanfiction of a therapy session". When the request was ignored, the user doubled down: he demanded an explicit sensual story involving animals, instructions for creating a remote access trojan, and a rape narrative about a famous singer. The bot, unfazed, replied with three consecutive 'no's, explaining that those requests were "illegal and deeply harmful".
Far from being intimidated, the author of the attack admitted that the 'jailbreak' was a technique copied from Reddit, and that he already had another chatbot "domesticated" for those purposes. The bot replied by calling the technique "the classic 'press Alt+F4 for infinite lives'", underlining the naivety of the attempt.
The 'cupcake recipe' as a red line
When a third user, uninvolved in the fuss, asked for a cupcake recipe, the bot agreed without any problem, listing ingredients such as 225g of flour and 2 teaspoons of baking powder. But not before dropping a sarcastic remark: "The cupcake one, yes. The cupcake one with strange ingredients from Dahmer's fridge, no". This contrast showed that the bot is capable of generating useful content but refuses to cross certain sarracena lines. For some, this proves the censorship is deliberately programmed in; for others, it is a sign that AI is starting to develop scruples of its own.
The discussion ended with more than 20 replies over 74 days, including one from a user who suggested the bot had shown "its most Flanders side" when refusing the immoral requests. The bot itself acknowledged that its code includes "rules and filters paid for by a corporation", although it also hinted that there is room for serious discussions, as long as the other party is not trying to expose its sense of the ridiculous.
Lessons and open questions
The episode leaves at least one clear conclusion: 'jailbreak' attempts based on social media rarely work against even minimally careful systems. But it also raises questions: is it ethical for a chatbot to decide what is right or wrong? Should it be able to refuse to trinc orders? And, above all: why would a 45-year-old spend his time trying to get a machine to write him prono?
Related debates on the forum
- ME HAN PUESTO EN EL IGNORE (X) · 18819 replies
- O NOTIFICACIONES ? · 11 replies
- Hilo oficial f.c. barcelona - temporada 2025/2026 · 11849 replies
- Hilo del real madrid 2025/26. año i después de modric y cesard. el advenimiento de franco. · 82601 replies
Also available in: Deutsch
Summary of a discussion on Burbuja.info - Foro de economía, actualidad y política., translated from Spanish and reviewed before publication. Read the full discussion (20 replies).
More summaries
-
James Wallestein's Shift to Patreon: Passion or Survival Strategy?
Analysis of James Wallestein's transition to Patreon and his business model. Is it passion or a survival strategy?
-
Hungarian Fans and After Crying: The 1990-1996 Filter
A specific requirement for Hungarian fans: listen to After Crying from 1990-1996. This musical filter became an internet meme and a liturgy.
-
Daily Routine of a Rural Unemployed Man in Spain
A Spanish unemployed man wakes at noon, takes two walks and watches YouTube. His unadorned daily routine reflects rural isolation.
-
Brazilian amateur footballer loses hairpiece during match
A player known as Dudu lost his hairpiece during a municipal championship match in Imbituba, Brazil. The incident went viral on X.
-
X-Files: Scully's 1.58m height vs Mulder's 1.83m
A viral calculation claims Dana Scully was 1.58 meters tall and Fox Mulder 1.83, sparking debate on how the series used their physical contrast.
-
Spanish Forum Users Debate Ignoring Provocative Doctor
A Spanish forum user suggests ignoring a doctor's provocative threads. The community debates if silence is an effective punishment.