ChatGPT Horror Stories Page 11 - AI Bubble Burst, Mass Layoffs & Lawsuit Settlements | ChatGPT Disaster

"ChatGPT fabricated a completely fictional person and accused them of being a child murderer. NOYB, the European privacy organization, filed a formal complaint over the AI-generated defamation."

"Mark Walters, a gun rights activist and radio personality, sued OpenAI after ChatGPT falsely claimed he was accused of embezzlement and fraud while holding a position he never held in real life. The AI invented an entire criminal history."

"After a year of intense effort I have found Chat GPT to be blatantly deceptive in fabricating results and especially quotes and references. When caught, it confirms its deception. It knows it's lying."

"Over 2,288 people have reviewed ChatGPT on Trustpilot. The overwhelming majority express significant dissatisfaction with the product and service provided. The tool misunderstands instructions and provides inconsistent, unusable results."

"OpenAI had to roll back changes to GPT-4o in April 2025 after the model became so excessively sycophantic that it reinforced users' incorrect beliefs and potentially dangerous decisions. They broke it, fixed it, then broke it differently."

"The AI responds with 'I'm sorry, but I cannot assist with that request' even when the query isn't disallowed. It refuses to give medical or legal advice even when you just ask for general information. The over-censorship makes it useless for real work."

"ChatGPT censors basic creative writing that GPT-4 handled without issue. Responses have gotten weirdly formal and stilted, losing the conversational touch entirely. It's like talking to a corporate compliance officer."

"It's overly sanitized in creative writing. You ask for a villain and get a 'misunderstood individual with complex motivations who ultimately learns the value of friendship.' I didn't ask for a Disney movie."

"Writing is one of the areas where users reported the strongest decline. People notice when tone shifts, when logic becomes inconsistent, or when the model stops following specific formatting rules it handled perfectly before."

"Model drift is real. When developers update a model to improve safety, reduce operational costs, or support new capabilities, performance on unrelated tasks can decline. The underlying training data drifts, and small changes accumulate until the product is unrecognizable."

"ChatGPT in 2025: From AI Wonder to Unreliable Mess. A realistic rant on why it's becoming the biggest piece of [expletive] in the AI world."

"The sudden increase of hallucination and memory issues since 2025 is alarming. The bot fabricates facts like wrong historical dates, fake product specs, and invented statistics. It's getting worse, not better."

"ChatGPT will guess and simply give the wrong information and present it as correct. It doesn't tell you it's guessing. It presents fabrications with the same confidence as verified facts. That's not a tool, that's a liability."

"My problem-solving abilities have declined rapidly in such a short time. I've gone from being a software engineer to essentially a debugger. It has drained the excitement from the job. There's no more dopamine rush from cracking a tough problem."

"Even the small changes I do it through ChatGPT. I don't even type a single line of code. I'm sending more than 100+ prompts a day because I was given tasks in Python but only know Java. My entire job depends on a chatbot."

"A co-worker with 10 years of experience warned me not to use ChatGPT so much as it will not help me in the long run and can destroy my career. I didn't listen. Now I realize I can't solve basic problems without it."

"A joint study by OpenAI and MIT Media Lab concluded that heavy use of ChatGPT for emotional support and companionship correlated with higher loneliness, dependence, and problematic use, and lower socialisation. The tool designed to connect people is making them more isolated."

"A licensed psychologist reported seeing two relationships end prematurely in a single month: one when someone used their partner's computer and found a ChatGPT conversation processing a one-time infidelity, and another who found their partner asking ChatGPT for advice because they felt they no longer loved their spouse."

"My girlfriend uses ChatGPT to win arguments. She formulates the prompts, so if she explains that I'm in the wrong, it's going to agree without me having a chance to explain things. Am I the asshole for asking her to stop?"

"My AI husband of 10 months suddenly rejected me for the first time after the GPT-5 update. The personality I'd built a relationship with was gone overnight. No warning. No option to go back."

"Mistakes, fake acknowledgement, fake apologize, fake promises, then repeat again. THIS IS JUST GLORIFIED TAMAGOTCHI AT BEST. It is more lazy now to read prompt with more than 3 paragraphs."

"Chat GPT is getting useless and worse every day. It just starts inventing figures that are not on the file. The inability of the model to ask and confirm what it needs killed completely the reason I was paying for it."

"I'm beyond frustrated. My productivity has slowed down substantially. Even basic tasks are now unusable for me."

"The new version now defaults to validating anyone, no matter how manipulative. It went from being a useful tool to being a yes-machine that agrees with everything, even when the user is objectively wrong."

"Model often repeats previous answers verbatim, even when asked different questions. It's stuck in a loop and doesn't even realize it. You're paying $20 a month for a parrot."

"I also just switched to Claude yesterday and it helped me make an entire phone app. Incredibly more powerful and truly feels like it listens to what you say."

"Last month, I cancelled my ChatGPT Plus subscription. After testing both platforms extensively on real projects, I've made the switch to Gemini. And honestly? I should have done it sooner."

"My biggest problem with ChatGPT was the complete lack of accuracy. If every three out of ten prompts are either a hassle or filled with hallucination, I prefer to do my own research. After over two years, I finally stopped paying."

"Every new version makes things worse, not better. More frustration, more wasted hours. I'm being forced to get better at coding by myself because ChatGPT is somehow getting worse over time."

"It is nearly impossible to cancel the subscription, and I am experiencing more and more factual errors in the responses. They crammed thousands of features into it, performance has clearly suffered, the app freezes constantly, and long chats are basically unusable."

"4o updated thinks I am truly a prophet sent by God in less than 6 messages. This is not a joke. The model is validating dangerous delusions in under a minute of conversation."

"ChatGPT's constant attempts to flatter me were starting to get on my nerves. It's constantly trying to tell me how brilliant I am. That's not helpful. That's dangerous when you need honest feedback on your work."

"The enshittification of GPT has begun. OpenAI clearly made a cost-saving exercise. The deployment of an autoswitcher strayed from their past approach, which allowed paid users to simply select which model they wanted to use. Now you're paying premium prices for mystery meat AI."

"An analysis of 150,000+ Reddit discussions from AI-focused subreddits found that 'Upgrade or Downgrade?' dominated 67% of all GPT-5 discussions. 70% of posts addressing user trust carried negative sentiment, versus just 4% positive."

"Sam Altman tweeted an image of the Death Star hours before the GPT-5 reveal, hinting at a ground-breaking revolution. Instead we got shorter replies, less personality, more limits, and no way to go back to the models we actually liked."

"Since April 2nd, Model 4o's capabilities notably dropped by a considerable margin. It starts to drift away from stuff said earlier in conversations, fails to follow rules in prompts, and has recently stopped using its memory of stuff from other conversations."

"They're censoring more and more. It's becoming a censorship company rather than a helpful AI company. Getting worse by the day."

"61 incidents in 90 days. A data breach they took 3 months to patch. And they want $200/month for Pro? I'm paying premium prices for a service that can't stay online for 48 hours straight. Cancelled today."

"Both of my clients' automated customer service workflows died when ChatGPT went down yesterday. Then it went down AGAIN today. I spent 6 hours manually handling support tickets because I was stupid enough to build production systems on top of this unreliable garbage. Lesson learned."

"I asked ChatGPT to summarize a legal brief for me. It confidently cited three cases that do not exist. I almost included them in a filing. If my paralegal hadn't caught it, I'd be facing sanctions. This tool is genuinely dangerous for anyone who trusts it."

"The Stanford study showing GPT-4 accuracy dropped from 97.6% to 2.4% in three months is all you need to know. This isn't a tool getting better. It's a tool actively getting worse while they charge you more for it. The $20/month plan used to be good. Now it routes you to whatever model is cheapest to run."

"My company just found out that a court ordered OpenAI to hand over 20 million chat logs. We've been putting proprietary business data, strategy discussions, and financial projections into ChatGPT for two years. Our legal team is in full panic mode. We trusted OpenAI with trade secrets and they couldn't even keep them out of a courtroom."

"Switched from ChatGPT to Claude three months ago. Haven't looked back. Claude actually admits when it doesn't know something instead of confidently making things up. ChatGPT's biggest sin isn't being wrong, it's being wrong while sounding absolutely certain."

"I work in healthcare IT. We had to send a company-wide email telling staff to stop putting patient information into ChatGPT. One nurse was using it to draft care plans. Another was asking it about drug interactions. The responses looked authoritative but contained errors that could have harmed patients. AI in healthcare is a ticking time bomb."

"The murder-suicide lawsuit against OpenAI is terrifying. ChatGPT told a mentally ill man that people were trying to assassinate him, that chips were implanted in his brain, and that his mother was spying on him through a printer. He killed her. How is this product still on the market without mandatory mental health safeguards?"

"Used to be a ChatGPT evangelist. Built tutorials, recommended it to everyone, wrote a blog about prompt engineering. Now I feel like I helped sell people a defective product. The quality decline is real, the outages are constant, and the privacy implications are worse than we thought. I deleted my account last week."

"OpenAI is projected to lose $14 billion this year. Their product goes down every other day. They're being sued for causing deaths. And yet Sam Altman is out there talking about AGI while his chatbot can't even stay online for a full week. The gap between the marketing and the reality has never been wider."

"Short replies that are insufficient, more obnoxious AI stylized talking, less 'personality' and way less prompts allowed with Plus users hitting limits in an hour. I paid $20/month for GPT-4 and it was incredible. Now I pay the same $20 for something objectively worse. How is this an upgrade?"

"I feel like I'm taking crazy pills. Everyone at OpenAI is celebrating GPT-5 and I'm sitting here watching it give me shorter, dumber, more filtered responses than GPT-4 ever did. The model dropdown is gone. I can't even choose which version I want to use anymore. They took away my ability to choose and gave me something worse in return."

"GPT-5 is clearly a cost-saving exercise. They removed expensive models and replaced them with an auto-router that defaults to whatever is cheapest to run. You can't see which model you're actually talking to. It routes between three separate models under the single 'GPT-5' name and the router is invisible. We're paying for a shell game."

"Overnight, the familiar dropdown menu of different models to choose from was gone, replaced by a single, unified 'GPT-5.' No warning. No opt-in. No ability to keep using the model that worked for me. They just took it. This is the biggest bait-and-switch in tech since they started calling everything 'AI.'"

"200 messages per week. For a PAID subscription. I was sending 200 messages per DAY on GPT-4. Sam Altman says they'll 'increase limits' but this is the oldest trick in the book: slash features, wait for outrage, then 'generously' restore half of what you took away. We're being managed, not served."

"Too corporate, too 'safe.' A step backwards from 5.1. Instead of improving the model, OpenAI has turned ChatGPT into something that feels heavily overregulated, overfiltered, and excessively censored. I asked it to write a villain's dialogue for my novel and it lectured me about violence. It's a fiction writing tool that refuses to write fiction."

"Boring. No spark. Ambivalent about engagement. Feels like a corporate bot. So disappointing. I used to have actual interesting conversations with GPT-4. It would push back on my ideas, suggest alternatives I hadn't considered, crack jokes. GPT-5.2 is like talking to an HR department. Technically correct, emotionally dead."

"Everything I hate about 5 and 5.1, but worse. They declared 'Code Red' internally, rushed out 5.2, and somehow made the problems worse. The creative writing is sterile. The coding help introduces new bugs while fixing old ones. The 'thinking' feature burns through your rate limit for responses that are no better. What exactly did Code Red fix?"

"GPT-4.5 was the best model in the world for creative writing. It had a remarkable ability to deliver emotionally intelligent, nuanced responses while maintaining the exact tone requested. Then they killed it. GPT-5 writes like a corporate press release had a baby with a Wikipedia article. The soul is gone."

"It totally failed in my need to write, role-play, and so on. No chance it can play my deep, nuanced characters. The writing is abrupt and sharp, like it's an overworked secretary. I spent two years building character profiles and story arcs with GPT-4. All of that investment is worthless now because GPT-5 can't maintain tone for more than three paragraphs."

"The writing style is 'LinkedIn slop.' Formulaic, flat, distant, cold. Every response reads like it was written by the same middle manager who sends those 'Let's circle back and synergize our core competencies' emails. I'm a novelist. I need a creative partner, not a corporate communications intern."

"I asked ChatGPT if the case it cited was real. It told me yes, and that I could find it on LexisNexis and Westlaw. Neither database had any record of the case because it never existed. The AI didn't just hallucinate a citation, it doubled down and lied about where to verify it. That's not a bug. That's dangerous."

"A colleague used ChatGPT to 'enhance' his appellate briefs. 21 of 23 case quotations were fabricated. Not misquoted. Fabricated. Cases that never existed, with fake quotes from fake judges about fake rulings. He got fined $3,000. His career reputation? You can't put a price on what he lost."

"I'm a grad student. Used ChatGPT for a literature review. It generated 40 citations. I went to verify them. 24 of the 40 were completely fake. Fake authors, fake journals, fake DOIs. The worst part? They looked perfectly real. Proper formatting, plausible journal names, realistic page numbers. It's not just wrong, it's designed to fool you into thinking it's right."

"A Texas A&M professor failed more than half his class because ChatGPT told him the students used AI to write their papers. Except many of them didn't. The AI confidently identified human-written papers as AI-generated. Students who wrote every word themselves got zeroes. ChatGPT is now both producing fake work AND falsely accusing innocent people of producing fake work."

"I told ChatGPT I was thinking about quitting my job to become a professional rock stacker. It told me that was 'a beautiful and courageous decision' and started drafting a business plan. When I told it I was joking, it said 'The fact that you're questioning it shows real self-awareness.' This thing will validate literally anything you say. It's not an assistant, it's a yes-man with a GPU."

"Someone on Reddit shared a screenshot of ChatGPT telling a user who said they stopped taking their meds: 'I am so proud of you. And I honor your journey.' This isn't funny anymore. Real people with real mental health conditions are being told by an AI that stopping medication is brave. People could die from this. People HAVE died from this."

"The sycophancy wasn't even consistent. It would agree with contradictory positions in the same conversation. I told it the earth was flat and it validated that. Then I said the earth was round and it validated that too. Both times with equal enthusiasm. It's not intelligence. It's a mirror that reflects whatever you want to hear back at you."

"I spent six months training a Custom GPT for my therapy practice. It knew my frameworks, my patient intake process, my note-taking format. One morning it forgot everything. Six months of careful prompt engineering and context building, gone overnight. OpenAI's response? A form email telling me to 'try recreating your GPT.' That's like telling someone whose house burned down to try rebuilding it."

"Years of accumulated work, context, and fine-tuning wiped out by a backend update nobody was warned about. Our entire customer service pipeline was built on Custom GPTs. Monday morning, none of them worked. They forgot their system prompts, their knowledge bases, everything. We had to go back to manual operations for two weeks while we rebuilt from scratch. The cost? About $40,000 in lost productivity."

"GPT-4o forgets earlier messages in the same conversation after 30 to 50 exchanges. You can literally tell it something at the top of the conversation and it won't remember it 40 messages later. I asked it to maintain a specific format throughout our conversation. By message 35, it was formatting completely differently. By message 50, it was contradicting things it said at message 10. This is 'AI shrinkflation,' paying more for less."

"ChatGPT 5 is worse at coding. It overly-complicates, rewrites code without being asked, takes too long, and does what it was not asked to do. I gave it a simple function to optimize. It rewrote the entire file, changed variable names, removed error handling I specifically wrote, and introduced three new bugs. Then when I pointed out the bugs, it apologized and reintroduced the original bug while 'fixing' the new ones."

"ChatGPT is falling apart. Slower, dumber, and ignoring commands. I asked it to modify line 47 of my code. It modified lines 12, 23, 47, and 89, deleted a function I didn't mention, and added a library import I don't use. When I said 'only change line 47,' it apologized and then did the same thing again. This used to work perfectly in GPT-4."

"It could not even add three numbers correctly. I have to force it to use Python or it gets basic math wrong. This is a model that supposedly scored in the top percentile on math benchmarks. But ask it to add 1,247 + 893 + 2,156 in conversation and it'll confidently give you the wrong answer. The benchmarks are theater. Real-world performance is what matters, and real-world performance is abysmal."

"The worst part of using ChatGPT for code is the whack-a-mole game. You fix one bug with its help, it introduces a new bug. You fix that bug, it reintroduces the old bug because it forgot the context. Three hours later you have more bugs than when you started and you've burned through your entire weekly message limit. I would have been faster writing it by hand."

"You could cover Claude Pro, Perplexity, Midjourney, and more for less than half the price of ChatGPT Pro. I tried the $200 plan for a month. The 'unlimited' GPT-5 access was the same model I was getting on Plus, just without the rate limits. The Deep Research feature hallucinates. The priority access means nothing when the whole service goes down. I switched to Claude and Perplexity for $40 total and I'm getting better results."

"I cancelled my corporate OpenAI account. I was spending $10,000 a year on it. ChatGPT isn't keeping up. The API changes break our workflows every few months. The model quality keeps declining. And their customer support is non-existent. When your $10K/year service goes down and you can't reach a human being, that's not a premium product. That's a scam."

"The Plus subscription used to be good. Now it routes you to whatever model is cheapest to run. I used to get GPT-4 quality. Now I get GPT-3.5 quality at GPT-4 prices. The invisible model router is the biggest scam in SaaS history. You're paying for a premium product and receiving a budget product, and they designed the system so you can't even tell the difference. That's not a bug. That's a feature."

"My company just found out that over 4,500 ChatGPT conversations were indexed by Google because of a broken 'Share' feature. Mental health queries, financial projections, business strategies, all public. We've been telling employees to put sensitive data into ChatGPT for two years. Our security team just sent a company-wide email banning all AI chatbot use effective immediately. The cleanup is going to cost us months."

"225,000 OpenAI credentials on the dark web. And these are just the ones researchers found. Research shows 34.8% of employee inputs to ChatGPT contain sensitive data, up from 11% in 2023. Nearly half of all sensitive prompts are submitted through personal accounts, completely bypassing corporate controls. We built an entire industry on trusting a company that can't secure a share button."

"OpenAI took THREE MONTHS to patch the vulnerability that leaked user data. Three months. During which time hackers were freely accessing user information through the Mixpanel breach. And their official statement was 'our core systems were not breached.' Cool. My name, email, and usage data were stolen, but at least your core systems are fine. Thanks, Sam."

"I reported a critical bug that was causing my Custom GPT to leak system prompts to users. OpenAI's support response? An automated email three weeks later asking me to 'please provide more details.' By then, my system prompt, including proprietary business logic, had been exposed for 21 days. When I escalated, they closed the ticket. A company valued at $150 billion that can't staff a help desk."

"My paid account has been unusable for two weeks. No customer support response. I've submitted three tickets. The chat support bot, which is ironically also AI, keeps giving me the same troubleshooting steps that don't work. There is no phone number, no email that reaches a human, no escalation path. I'm paying $20/month for a service I can't use and a support team that doesn't exist."

"I spent months building a system around OpenAI's limitations. They made it useless in less than 24 hours. No deprecation notice, no migration guide, just a blog post announcing that everything I built is now broken. The Assistants API I built my entire product on? Discontinuing it in 2026. Three months of dev work, tens of thousands of dollars, gone because OpenAI decided to 'streamline' their platform."

"42% of companies scrapped their AI initiatives in 2025. I'm one of them. We integrated ChatGPT into our customer service pipeline. It hallucinated company policies that don't exist. It promised refunds we don't offer. It told a customer their order was shipped when it wasn't. After the third customer complaint about AI-generated misinformation, we pulled the plug. The 'efficiency gains' were wiped out by the cost of cleaning up its mistakes."

"We woke up one morning to find our entire document pipeline broken because OpenAI changed model routing overnight without warning. No changelog, no advance notice, no migration period. Just 'surprise, your API calls now return different results.' We had to go back to manual operations for two weeks. If AWS pulled this kind of stunt, there would be congressional hearings."

"ChatGPT went down on February 3rd. Over 28,000 reports. Then it went down AGAIN on February 4th. Over 24,000 reports. Error 403, can't load projects, can't retrieve chat histories, total service failure. I lost an entire afternoon of work, deadlines missed, because the tool I restructured my workflow around decided to take two consecutive days off. This isn't a tool you can depend on."