AI Classifier · Image: OpenAI (press image)
Archivetitankilledai content media

AI Classifier

AI Classifier, OpenAI's experimental web tool for detecting AI-generated text, was shut down on July 20, 2023, after six months of operation.

In our view, releasing a text detector that was wrong nearly three-quarters of the time proved that even OpenAI could not solve the problem created by its own chatbot.

Our take
Cause:model capability gapsolution seeking problem

OpenAI launched AI Classifier in January 2023 to detect AI-generated text, but quietly discontinued the free tool on July 20, 2023, due to an unacceptably low rate of accuracy.

Fail ScoreOur take
34/100
Hype 10/25
Money 2/25
Spectacle 8/25
Lesson 14/25
rubric
Hype · our take
3/5 promised → 1/5 delivered
Lifespan
200 days
Timing · our take
unclear
Dates
announced · shipped · discontinued
Where
Incumbent Product · Internal
manyfails.com/fails/ai-classifier
Download cardPost to X

Product and company names are used for identification only.

The Pitch

OpenAI announced AI Classifier at the start of the year as a dedicated mechanism to identify machine-written text E4E4E4. The tool was trained to distinguish between human and artificial intelligence writing across various model providers E4E4E4. Users interacted with the free utility through a web form where they pasted a passage of text into a box to receive an automated likelihood score E4. The tool was intended for educators and researchers seeking to identify automated plagiarism or academic dishonesty E4E4E4. At launch, OpenAI warned that the classifier performed poorly on submissions under one thousand words and failed on coding and non-English text E4. Internal evaluations demonstrated that the classifier correctly identified only twenty-six percent of AI-written text and generated a nine percent false positive rate on human prose E4.

What Happened

OpenAI rolled out the AI Classifier in January 2023 in response to concerns regarding academic integrity and generative text proliferation E4E4E4. The public reception quickly soured because the software was wrong too often and struggled to classify short or non-English passages reliably E4E4. Users found that the tool mislabeled longer text, frequently generated incorrect classifications, and proved ineffective for catching student plagiarism E4. Controlled benchmarks confirmed that the classifier successfully detected AI text only twenty-six percent of the time while falsely accusing human writers in nine percent of trials E4. On July 20, 2023, OpenAI discontinued the service and removed active links to the web application without a public press release E4E2E4E4E4E4. The company updated an existing blog post to state that the classifier was retired due to its persistent low rate of accuracy E2. OpenAI stated that it would incorporate user feedback and research more effective provenance techniques for text, visual, and audio media E2. The sudden shutdown left educators worldwide disappointed as viable alternatives for catching AI-generated submissions remained elusive E4E4.

Why It FailedOur take

In our view, the failure of AI Classifier was driven by an insurmountable model capability gap. Post-hoc statistical classification cannot reliably separate human writing from model-generated prose when generative architectures emulate human linguistic patterns. Detecting text with a twenty-six percent true positive rate meant three-quarters of machine-written submissions escaped detection, while a nine percent false positive rate imposed unacceptable reputational risks on innocent students. Furthermore, OpenAI pursued a solution seeking a problem by shipping a probabilistic detector without the cryptographic provenance or watermarking required for reliable verification.

Lessons and What Would Have Worked

  1. In our view, founders should not release public verification tools when false positive rates impose severe reputational or disciplinary penalties on innocent users.
  2. If a detection model catches barely a quarter of target outputs, shipping it publicly diminishes trust faster than offering no solution at all.
  3. In our opinion, content authenticity requires cryptographic watermarking at generation time rather than post-hoc statistical classification.

Timeline

  1. OpenAI rolls out the AI Classifier tool to detect content generated by artificial intelligence.E4
  2. OpenAI quietly discontinues AI Classifier due to its low accuracy rate.E2

What happened to AI Classifier?

What was AI Classifier?
OpenAI launched AI Classifier in January 2023 to detect AI-generated text, but quietly discontinued the free tool on July 20, 2023, due to an unacceptably low rate of accuracy.
What happened to AI Classifier?
OpenAI rolled out the AI Classifier in January 2023 in response to concerns regarding academic integrity and generative text proliferation .
Why did AI Classifier fail?
In our view, the failure of AI Classifier was driven by an insurmountable model capability gap.

Receipts

  1. E4
    AI Detector Fail: This week Open AI quietly shut down its new AI Classifier due to accuracy | Generative Artificial Intelligence, Reference, & Instruction Discussion Group (GAIR&I)

    connect.ala.org · · Archived

    supports: shutdown, root_cause, aftermath

    Supporting quote
    As of July 20, 2023, the AI classifier is no longer available due to its low rate of accuracy. We are working to incorporate feedback and are currently researching more effective provenance techniques for text, and have made a commitment to develop and deploy mechanisms that enable users to understand if audio or visual content is AI-generated.

    Accessed

  2. E2
    OpenAI scuttles AI-written text detector over 'low rate of accuracy' | TechCrunch

    techcrunch.com · · Archived

    supports: shutdown, shutdown-reason · single source, provenance-research · single source

    Supporting quote
    OpenAI has shut down its AI classifier, a tool that claimed to determine the likelihood a text passage was written by another AI. While many used and perhaps unwisely relied on it to catch low-effort cheats, OpenAI has retired it over its widely criticized “low rate of accuracy.”

    Accessed

  3. E4
    OpenAI Shuts Down ChatGPT Plagiarism Detector Because It Doesn’t Work | Observer

    observer.com · · Archived

    supports: shutdown, model_accuracy · single source, pitch

    Supporting quote
    As of July 20, OpenAI has quietly pulled the plug on their A.I. detection software, AI Classifier, citing its “low rate of accuracy.” If this is the first time you’re hearing of this, don’t be surprised—OpenAI buried the news so deep underground that it took a week for anyone to even notice it was there.

    Accessed

  4. E4
    OpenAI shuts down its AI detection tool

    computing.co.uk · · Archived

    supports: shutdown, announcement, capabilities, root_cause, user_model · single source, warning_sign · single source

    Supporting quote
    GPT-4 creator OpenAI has taken down links to its AI Classifier tool. The AI detection tool was announced at the start of the year, prompted by concerns about plagiarism, academic integrity, and the potential for generative AI to generate misinformation.

    Accessed

  5. E4
    OpenAI shuts down AI Classifier, its tool to detect AI-generated content

    indiaai.gov.in · · Archived

    supports: shutdown, announcement, capabilities, limitations · single source, aftermath

    Supporting quote
    OpenAI said in a blog post, “As of July 20, 2023, the AI classifier is no longer available due to its low accuracy rate. We are working to incorporate feedback and researching more effective provenance techniques for text. We have committed to developing and deploying mechanisms enabling users to understand whether audio or visual content is AI-generated.”

    Accessed

  6. E4
    AI Classifier - Intelligent Email Organization for Gmail & Workspace

    aiclassifier.tech · · Archived

    supports: pitch, real_time_processing · single source, custom_categories · single source, integration · single source, business_model · single source, prices · single source, users · single source

    Supporting quote
    Stop drowning in your inbox. AI Classifier automatically categorizes, prioritizes, and organizes your Gmail emails using advanced artificial intelligence. Save 3+ hours per week and never miss important communications again. [Start Free Trial\ →](https://app.aiclassifier.tech/) [Watch Demo\ ▶](https://aiclassifier.tech/#demo) -97%Confidence 3hrsSaved/Week 10k+Users ✏️ Compose 📥Inbox347 ⭐Starred Labels URGENT12 Finances28 Important Docs7 Client Projects45 Subscriptions89 ↻ 1-50 of 347 ☆ Sarah ChenURGENTQ4 Strategy Review - Action Required 10:24 AM ☆ Bank of AmericaFinancesYour November statement is ready 9:47 AM ☆ DocuSignImportant DocsPlease review and sign: NDA Agreement Yesterday ☆ Michael from TechCorpClient ProjectsProject timeline update and next steps Nov 20 ☆ NetflixSubscriptionsNew releases this week you might enjoy Nov 20 🧠AI Classifier ## Powerful Features for Email Management Everything you need to transform your inbox from chaos to clarity ### AI-Powered Classification Advanced machine learning algorithms automatically categorize emails with confidence-aware classification, learning from your preferences. ### Real-time Processing Emails are classified instantly as they arrive. No delays, no manual work - just perfectly organized messages. ### Smart Prioritization Important emails are automatically flagged and moved to priority folders, ensuring you never miss critical communications. ### Analytics Dashboard Detailed insights into your email patterns, response times, and productivity metrics with beautiful visualizations. ### Custom Categories Create up to 10 custom classification categories tailored to your specific workflow and business needs. ### Seamless Integration Works perfectly with Gmail and Google Workspace. No disruption to your existing email workflow. ## How AI Classifier Works Get started in minutes with our simple 3-step process 1 ### Connect Your Gmail Securely connect your Gmail or Google Workspace account with just a few clicks. We only access what's needed for classification. 2 ### Configure Your Categories Customize classification categories and rules to match your business needs. Configure AI settings, priority levels, and automation preferences. 3 ### Enjoy Organized Inbox New emails are automatically classified and organized. Your inbox transforms from chaos to clarity. ## Simple, Volume-Based Pricing Choose the plan that matches your email volume. No hidden fees, no surprises. 14-Day Free Trial ### Free Trial R0/14 days Up to 2,000 emails - ✅ All AI classification features - ✅ Up to 10 custom categories - ✅ Email support - ✅ No credit card required - ✅ **Free setup assistance included** [Start Free Trial](https://app.aiclassifier.tech/?plan=trial) Essential AI organization ### Starter R169/month Up to 2,000 emails/month - ✅ All AI classification features - ✅ Up to 10 custom categories - ✅ Email support - ✅ Gmail & Workspace integration [Choose Starter](https://app.aiclassifier.tech/?plan=starter) ⭐ Most Popular ### Personal R349/month Up to 5,000 emails/month - ✅ All AI classification features - ✅ Up to 10 custom categories - ✅ Email support - ✅ Gmail & Workspace integration [Start Personal Plan](https://app.aiclassifier.tech/?plan=personal) High-volume processing ### Professional R899/month Up to 15,000 emails/month - ✅ All AI classification features - ✅ Up to 10 custom categories - ✅ Email support - ✅ Gmail & Workspace integration [Choose Professional](https://app.aiclassifier.tech/?plan=professional) Enterprise-scale volumes ### Business R2699/month Up to 50,000 emails/month - ✅ All AI classification features - ✅ Up to 10 custom categories - ✅ Email support - ✅ Gmail & Workspace integration [Choose Business](https://app.aiclassifier.tech/?plan=business)

    Accessed

  7. E4
    OpenAI Shuts Down Flawed AI Detector

    searchenginejournal.com · · Archived

    supports: shutdown, capabilities, announcement

    Supporting quote
    OpenAI has discontinued its AI classifier, a tool designed to identify AI-generated text, following criticism over its accuracy. The termination was subtly announced via an update to an existing blog post.

    Accessed

Sentiment

At launch

No launch coverage quote recorded.

At death

No death coverage quote recorded.

No sentiment quotes recorded.

Last updated · Phil · How we source entries

Auto-published from sourced research; editorial review pending

Changelog

  • Published AI Classifier.
Corrections and right of reply

The Post-Mortem

One fail, the week’s deaths, watchlist moves and the Counter. Every Friday.

Confirm by email. No spam. Unsubscribing works, unlike the products we cover.

Two independent sources per claim. Opinion labelled. Corrections public.