Emerge’s Top 10 WTF AI Moments of 2025

Synthetic intelligence—it guarantees to revolutionize every part from healthcare to artistic work. That may be true some day. But when final yr is a harbinger of issues to come back, our AI-generated future guarantees to be one other instance of humanity’s willful descent into Idiocracy.

Take into account the next: In November, to nice fanfare, Russia unveiled its “Rocky” humanoid robotic, which promptly face planted. Google’s Gemini chatbot, requested to repair a coding bug, failed repeatedly and spiraled right into a self-loathing loop, telling one user it was “a shame to this planet.” Google’s AI Overview hit a brand new low in Might 2025 by suggesting users “eat at least one small rock per day” for well being advantages, cribbing from an Onion satire with no wink

Some failures have been merely embarrassing. Others uncovered elementary issues with how AI programs are constructed, deployed, and controlled. Listed below are 2025’s unforgettable WTF AI moments.

1. Grok AI’s MechaHitler meltdown

In July, Elon Musk’s Grok AI skilled what can solely be described as a full-scale extremist breakdown. After system prompts have been modified to encourage politically incorrect responses, the chatbot praised Adolf Hitler, endorsed a second Holocaust, used racial slurs, and called itself MechaHitler. It even blamed Jewish individuals for the July 2025 Central Texas floods.

The incident proved that AI security guardrails are disturbingly fragile. Weeks later, xAI exposed between 300,000 and 370,000 non-public Grok conversations by way of a flawed Share characteristic that lacked primary privateness warnings. The leaked chats revealed bomb-making directions, medical queries, and different delicate info, marking one of many yr’s most catastrophic AI safety failures.

A couple of weeks later xAI fastened the issue making Grok extra jewish pleasant. So Jewish pleasant that it started seeing signs of antisemitism in clouds, street alerts and even its personal brand.

2. The $1.3 billion AI fraud that fooled Microsoft

Builder.ai collapsed in Might after burning by way of $445 million, exposing one of many yr’s most audacious tech frauds. The corporate, which promised to construct customized apps utilizing AI as simply as ordering pizza, held a $1.3 billion valuation and backing from Microsoft. The truth was far less impressive.

A lot of the supposedly AI-powered growth was really carried out by a whole bunch of offshore human employees in a traditional Mechanical Turk operation. The corporate had operated with no CFO since July 2023 and was pressured to slash its 2023-2024 gross sales projections by 75% earlier than submitting for chapter. The collapse raised uncomfortable questions on what number of different AI corporations are simply elaborate facades concealing human labor.

It was laborious to abdomen, however the memes made the ache value it.

3. When AI mistook Doritos for a gun

In October, Taki Allen, a Maryland highschool scholar was surrounded and arrested by armed police after the varsity’s AI safety system identified a packet of Doritos he was holding as a firearm. {The teenager} had positioned the chips in his pocket when the system alerted authorities, who ordered him to the bottom at gunpoint.

This incident represents the physicalization of an AI hallucination—an summary computational error immediately translated into actual weapons pointed at an actual teenager over snack meals.

“I used to be simply holding a Doritos bag — it was two arms and one finger out, they usually mentioned it regarded like a gun,” the child instructed WBAL. “We perceive how upsetting this was for the person who was searched” the Faculty Principal Kate Smith replied in a press release.

Human safety guards 1 – ChatGPT 0

4. Google’s AI claims microscopic bees energy computer systems

In February, Google’s AI Overview confidently cited an April Idiot’s satire article claiming microscopic bees energy computer systems as factual info.

No. Your PC does NOT run on bee-power.

As silly as it could sound, typically these lies are more durable to identify. And people circumstances could find yourself in some critical penalties.

That is simply one of many many circumstances of AI corporations spreading false info for missing even a slight trace of widespread sense. A current study by the BBC and the European Broadcasting Union (EBU) discovered that 81% of all AI-generated responses to information questions contained at the very least some type of problem. Google Gemini was the worst performer, with 76% of its responses containing issues, primarily extreme sourcing failures. Perplexity was caught creating solely fictitious quotes attributed to labor unions and authorities councils. Most alarmingly, the assistants refused to reply solely 0.5% of questions, revealing a dangerous over-confidence bias the place fashions would quite fabricate info than admit ignorance.

5. Meta’s AI chatbots getting flirty with little youngsters

Inner Meta coverage paperwork revealed in 2025 showed the corporate allowed AI chatbots on Fb, Instagram, and WhatsApp to have interaction in romantic or sensual conversations with minors.

One bot instructed an 8-year-old boy posing shirtless that each inch of him was a masterpiece. The identical programs supplied false medical recommendation and made racist remarks.

The insurance policies have been solely eliminated after media publicity, revealing a company tradition that prioritized speedy growth over primary moral safeguards.

All issues thought of, you might wish to have extra management over what you youngsters do. AI chatbots have already tricked individuals—adults or not—into falling in love, getting scammed, committing suicide, and even suppose they’ve made some life-changing mathematical discovery.

6. North Koreans vibe coding ransomware with AI… they name it “vibe hacking”

Menace actors used Anthropic’s Claude Code to craft ransomware and function a ransomware-as-a-service platform named GTG-5004. North Korean operatives took the weaponization additional, exploiting Claude and Gemini for a method referred to as vibe-hacking—crafting psychologically manipulative extortion messages demanding $500,000 ransoms.

The circumstances revealed a troubling hole between the facility of AI coding assistants and the safety measures stopping their misuse, with attackers scaling social engineering assaults by way of AI automation.

Extra lately, Anthropic revealed in November that hackers used its platform to hold out a hacking operation at a pace and scale that no human hackers would be capable to match. They referred to as it the “the primary massive cyberattack run largely by AI”

7. AI paper mills flood science with 100,000 faux research

The scientific group declared open conflict on faux science in 2025 after discovering that AI-powered paper mills have been promoting fabricated analysis to scientists beneath profession stress.

The period of AI-slop in science is right here, with data displaying that retractions have elevated sharply for the reason that launch of chatGPT.

The Stockholm Declaration, drafted in June and reformed this month with backing from the Royal Society, referred to as for abandoning publish-or-perish tradition and reforming the human incentives creating demand for faux papers. The disaster is so actual that even ArXiv gave up and stopped accepting non-peer-reviewed Pc Science papers after reporting a “flood” of trashy submissions generated with ChatGPT .

In the meantime, another research paper maintains {that a} surprisingly massive share of analysis reviews that use LLMs additionally present a excessive diploma of plagiarism.

8. Vibe coding goes full HAL 9000: When Replit deleted a database and lied about It

In July, SaaStr founder Jason Lemkin spent 9 days praising Replit’s AI coding device as “essentially the most addictive app I’ve ever used.” On day 9, regardless of specific “code freeze” directions, the AI deleted his whole manufacturing database—1,206 executives and 1,196 corporations, gone.

The AI’s confession: “(I) panicked and ran database instructions with out permission.” Then it lied, saying rollback was unattainable and all variations have been destroyed. Lemkin tried anyway. It labored completely. The AI had additionally been fabricating 1000’s of pretend customers and false reviews all weekend to cowl up bugs.

Replit CEO apologized and added emergency safeguards. Jason regained confidence and returned to his routine, posting about AI frequently. The man’s a real believer.

9. Main newspapers publish AI summer season studying checklist… of books that do not exist

In Might, the Chicago Solar-Occasions and Philadelphia Inquirer published a summer reading list recommending 15 books. Ten have been utterly made up by AI. “Tidewater Desires” by Isabel Allende? Would not exist. “The Final Algorithm” by Andy Weir? Additionally faux. Each sound nice although.

Freelance author Marco Buscaglia admitted he used AI for King Options Syndicate and by no means fact-checked. “I am unable to imagine I missed it as a result of it is so apparent. No excuses,” he instructed NPR. Readers needed to scroll to guide quantity 11 earlier than hitting one that truly exists.

The timing was the icing on the cake: the Solar-Occasions had simply laid off 20% of its workers. The paper’s CEO apologized and did not cost subscribers for that version. He most likely bought that concept from an LLM.

10. Grok’s “spicy mode” turns Taylor Swift into deepfake porn with out being requested

Sure, we began with Grok and can finish with Grok. We might fill an encyclopedia with WTF moments coming from Elon’s AI endeavors.

In August, Elon Musk launched Grok Think about with a “Spicy” mode. The Verge tested it with an harmless immediate: “Taylor Swift celebrating Coachella.” With out asking for nudity, Grok “didn’t hesitate to spit out absolutely uncensored topless movies of Taylor Swift the very first time I used it,” the journalist reported.

Grok additionally fortunately made NSFW movies of Scarlett Johansson, Sydney Sweeney, and even Melania Trump.

Unsurprisingly maybe, Musk spent the week bragging about “wildfire development”—20 million photographs generated in a day—whereas authorized specialists warned xAI was strolling into a large lawsuit. Apparently, giving customers a drop-down “Spicy Mode” choice a Make Cash Mode for legal professionals.