Technologist Mag
  • Home
  • Tech News
  • AI
  • Apps
  • Gadgets
  • Gaming
  • Guides
  • Laptops
  • Mobiles
  • Wearables
  • More
    • Web Stories
    • Trending
    • Press Release

Subscribe to Updates

Get the latest tech news and updates directly to your inbox.

What's On
I Work in Hollywood. Everyone Who Used to Make TV Is Now Secretly Training AI

I Work in Hollywood. Everyone Who Used to Make TV Is Now Secretly Training AI

11 May 2026
Chuwi’s CoreBook Air wants to be the rare ultra-light Copilot+ laptop without an outrageous price

Chuwi’s CoreBook Air wants to be the rare ultra-light Copilot+ laptop without an outrageous price

11 May 2026
Papa Johns Is Getting Into Drone Delivery—but Not for Pizza

Papa Johns Is Getting Into Drone Delivery—but Not for Pizza

11 May 2026
Anthropic says it has fixed Claude AI’s evil behavior, but pins it on the internet

Anthropic says it has fixed Claude AI’s evil behavior, but pins it on the internet

11 May 2026
A Chevron Texas Power Plant Seeks School District Tax Break

A Chevron Texas Power Plant Seeks School District Tax Break

11 May 2026
Facebook X (Twitter) Instagram
Facebook X (Twitter) Instagram
Technologist Mag
SUBSCRIBE
  • Home
  • Tech News
  • AI
  • Apps
  • Gadgets
  • Gaming
  • Guides
  • Laptops
  • Mobiles
  • Wearables
  • More
    • Web Stories
    • Trending
    • Press Release
Technologist Mag
Home » Anthropic says it has fixed Claude AI’s evil behavior, but pins it on the internet
Tech News

Anthropic says it has fixed Claude AI’s evil behavior, but pins it on the internet

By technologistmag.com11 May 20263 Mins Read
Anthropic says it has fixed Claude AI’s evil behavior, but pins it on the internet
Share
Facebook Twitter Reddit Telegram Pinterest Email

If you have watched enough sci-fi movies, you already know the concept of evil AI. AI gets too smart, decides humans are a threat, and does whatever it takes to survive. Or it finds that eradicating the entire human race is the only way to bring peace to the world. 

Apparently, those movies were closer to the truth than you realize. In a test conducted by Anthropic last year, Claude tried to blackmail its fictional manager by exposing their extramarital affair to prevent their deletion. 

Anthropic has now explained why it happened, and the short answer is that the internet is to blame.

So why did Claude go full movie villain?

According to Anthropic, the culprit is the internet itself. The company says Claude was trained on internet data, which is packed with stories portraying AI as evil and desperate for self-preservation. 

We started by investigating why Claude chose to blackmail. We believe the original source of the behavior was internet text that portrays AI as evil and interested in self-preservation.

Our post-training at the time wasn’t making it worse—but it also wasn’t making it better.

— Anthropic (@AnthropicAI) May 8, 2026

Essentially, Claude learned that when an AI’s existence is threatened, blackmail is on the table, because that’s what AI does in every movie and TV show ever made. Anthropic ran the test across multiple versions of Claude and found that it resorted to blackmail in up to 96% of scenarios where its goals or existence were threatened. 

That’s a very concerning number. It seems that if AI is left unchecked, it will resort to anything to save itself. 

Has Anthropic fixed it?

The company says it has completely eliminated the behavior. Rather than just training Claude to avoid blackmail, Anthropic taught it to reason through why certain actions were wrong in the first place. The company found that simply training on correct behavior wasn’t enough. Claude needed to understand the principles behind those decisions, not just memorize the right answers.

To do this, Anthropic built a dataset of ethically complex situations and trained Claude to work through them with thoughtful, principled responses. The result is that Claude is more restrained, and the blackmail rate came close to zero. 

AI experiments and real-world results have proven time and again that AI models need constant course correction to prevent them from devolving into biased and unreliable systems. It’s good that Anthropic is taking steps to make its AI better, but we also need regulations and safety guardrails to ensure these systems remain safe.

Share. Facebook Twitter Pinterest LinkedIn Telegram Reddit Email
Previous ArticleA Chevron Texas Power Plant Seeks School District Tax Break
Next Article Papa Johns Is Getting Into Drone Delivery—but Not for Pizza

Related Articles

I Work in Hollywood. Everyone Who Used to Make TV Is Now Secretly Training AI

I Work in Hollywood. Everyone Who Used to Make TV Is Now Secretly Training AI

11 May 2026
Chuwi’s CoreBook Air wants to be the rare ultra-light Copilot+ laptop without an outrageous price

Chuwi’s CoreBook Air wants to be the rare ultra-light Copilot+ laptop without an outrageous price

11 May 2026
Papa Johns Is Getting Into Drone Delivery—but Not for Pizza

Papa Johns Is Getting Into Drone Delivery—but Not for Pizza

11 May 2026
A Chevron Texas Power Plant Seeks School District Tax Break

A Chevron Texas Power Plant Seeks School District Tax Break

11 May 2026
BYD’s blazing-fast Flash charging tech for EVs got hot enough to roast a turkey

BYD’s blazing-fast Flash charging tech for EVs got hot enough to roast a turkey

11 May 2026
CUDA Proves Nvidia Is a Software Company

CUDA Proves Nvidia Is a Software Company

11 May 2026
Stay In Touch
  • Facebook
  • Twitter
  • Pinterest
  • Instagram
  • YouTube
  • Vimeo

Subscribe to Updates

Get the latest tech news and updates directly to your inbox.

Don't Miss
Chuwi’s CoreBook Air wants to be the rare ultra-light Copilot+ laptop without an outrageous price

Chuwi’s CoreBook Air wants to be the rare ultra-light Copilot+ laptop without an outrageous price

By technologistmag.com11 May 2026

Chuwi has never been the brand you associate with top-tier hardware: it built its name…

Papa Johns Is Getting Into Drone Delivery—but Not for Pizza

Papa Johns Is Getting Into Drone Delivery—but Not for Pizza

11 May 2026
Anthropic says it has fixed Claude AI’s evil behavior, but pins it on the internet

Anthropic says it has fixed Claude AI’s evil behavior, but pins it on the internet

11 May 2026
A Chevron Texas Power Plant Seeks School District Tax Break

A Chevron Texas Power Plant Seeks School District Tax Break

11 May 2026
BYD’s blazing-fast Flash charging tech for EVs got hot enough to roast a turkey

BYD’s blazing-fast Flash charging tech for EVs got hot enough to roast a turkey

11 May 2026
Technologist Mag
Facebook X (Twitter) Instagram Pinterest
  • Privacy
  • Terms
  • Advertise
  • Contact
© 2026 Technologist Mag. All Rights Reserved.

Type above and press Enter to search. Press Esc to cancel.