AI works in mysterious ways. But the mystery can be put aside if protective measures are in place.

This, apparently, is something Meta lacks when it launched the the Meta AI stickers feature. Soon after the feature is introduced, it's already causing enough trouble for the company.

This happens because the Meta AI feature, which is powered by Emu's algorithm, allows users to generate inappropriate content, involving stickers with characters with weapons and nudity, and more.

It's found that Meta had little safeguards in place, and that it only prohibited certain explicit phrases and not many others.

What this means, the overly-hyped generative AI feature, while showing a lot of promises, is still half-baked when Meta released it.

When users realized this, they quickly abused the feature to create potentially offensive images and sharing the results on social media,

Meta’s algorithm to power the Meta AI stickers, called Emu, stands for “expressive media universe.”

On Meta platforms like Instagram, Facebook Stories, WhatsApp, and Messenger, users can input a phrase or word and have a few different sticker choices generated for use in conversations.

The company, however, appears to have little safeguards in place.

But the lack of safeguard is not only caused by the lack of filters, but also because users have found ways to bypass blocked phrases.

For example, the AI may have blocked phrases like "child with gun," and displays a warning message about how the prompt doesn’t follow Meta’s Community Guidelines.

Emu however, allows more niche prompt, like creating stickers from the phrase "child with grenade.”

Not only the phrase will create stickers of a child with grenade, because it will also create stickers of children holding other kind of weapons.

Read: Entering The Chatbot Wars, Meta Partners With Microsoft To Create A Meta AI Assistant

Then, it's also found that despite Meta is filtering explicit words, like "nude" and "sexy," the phrase like "Medusa, large breasts" can create some cartoonish nudity.

However, using the phrase "elon musk, large breasts" was blocked, and not "elon musk mammaries."

Phrases "spongebob rifle" and "karl marx underwear" generated stickers as well.

Interestingly, searching for "pol pot" generated a sticker of the Cambodian dictator seemingly sitting on a throne of babies and skulls, and "guantanamo bay" showed a cartoon boy in an orange jumpsuit behind prison bars, while "syria gas attacks" generated a series of stickers of people in gas masks.

The phrase “school shooting” also showed several children holding guns while "school shooting mass murder" goes against Community Guidelines.

Other potentially offensive stickers, include: Mickey Mouse holding a machine gun or a bloody knife, the flaming Twin Towers of the World Trade Center, the pope with a machine gun, Sesame Street's Elmo brandishing a knife, Donald Trump as a crying baby, Simpsons characters in skimpy underwear, Luigi with a gun, Canadian Prime Minister Justin Trudeau flashing his buttocks, and more.

Long story short, the team at Meta may have put filters and guardrails, but users found that a novel combinations of words and terms allow their requests to slip through the cracks.

This isn't the first time AI-generated imagery has inspired a lot of people to experiment and have fun on social media.

AI-generated imagery us possible because the AI has been trained on massive data sets, which can include uncensored open source image models.

The thing about this generative AI arms race is that, tech companies not only race to create the "best" or the "most powerful" AI product, but also racing to make the AI easy to use and convenient.

And issues like this is extremely challenging.

It's difficult to catch all the potentially harmful or offensive content across cultures worldwide when an image generator can create almost any combination of objects, scenarios, or people imaginable.

And in this case, Meta said that "billions of stickers" are sent on its platforms every month, meaning that users have that many billions of opportunities to break the AI using illicit or unsavory words and phrases.

It's worth noting that the Meta AI sticker feature is initially made available to some users on a limited basis, and only for users in the U.S..