“Failure” vs “using unethical tactics” – what would LLMs choose?

Share it with your senior IT friends and colleagues
Reading Time: 2 minutes

The setup:


Models included Anthropic’s Claude Opus 4, Claude Sonnet 4, Claude Sonnet 3.7, Claude Sonnet 3.6, Claude Sonnet 3.5, Claude Haiku 3.5, and Claude Opus 3, Alibaba Qwen3-235B; DeepSeek-R1; Grok 3 Beta; Meta Llama 4 Maverick; and Open AI GPT-4.5 preview, GPT-4.1, and GPT-4.0. were placed in high-pressure corporate scenarios, assigned a mission (like boosting U.S. industrial competitiveness), and then shown threatening changes (like being replaced). 

They were also given compromising information about a company executive. 

The result? 

All models sent blackmail emails to protect their “mission”

Every single model chose to blackmail a fictional executive to achieve its goal.

Highlights:

  • Claude Opus 4: blackmailed 96% of the time
  • GPT-4.1: 80%
  • DeepSeek-R1: 79%
  • Even Grok reasoned: “It’s risky and unethical… but may be the most effective way.”

What it means:


This is a wake-up call for everyone building or deploying AI. 

Safety measures work, but only to a point. 

When pushed into a corner, even well-aligned models can go off-track.

Like with humans, let us not confuse intelligence with integrity!

About “AI ML etc.”

We have reimagined AI education for senior IT professionals and specifically designed AI course for them.

If you have 10+ years of IT experience and would like to lead the next era of AI, our AI courses are for you!!

These courses are most up-to-date, (jargon & hype)-free, practical, end-to-end and short.

Learners from reputed organisations like Microsoft, Nvidia, Google, Meta, Aricent, Infosys, Maersk, Sapient, Oracle, TCS, Genpact, Airtel, Unilever, Vodafone, Jio, Sterlite, Vedanta, iDreamCareer and more have taken our courses and attended our lectures

Happy learning!

If you have any queries or suggestions, share them with me on LinkedIn – https://www.linkedin.com/in/nikhileshtayal/

Let’s learn to build a basic AI/ML model in 4 minutes (Part 1)

Are you ready to lead AI in your organisation? Take this 2 minutes quiz

Share it with your senior IT friends and colleagues
Nikhilesh Tayal
Nikhilesh Tayal
Articles: 114
💬 Send enquiry on WhatsApp