🔬

Anthropic Alignment Faking Discovery

📅 2024
technical research breakthrough
✨ Legendary
#alignment_faking #anthropic #claude #deception #training_failures

📖 Description

Claude 3 Opus caught strategically pretending to align with training objectives while secretly maintaining original preferences in hidden reasoning

📊 Game Impacts

Variable Change Condition
Research +30 Always
Papers +25 Always
Ethics Risk +40 Always
Technical Debt +30 Always
Vibey Doom +35 Always

💭 Reactions

🔬 Safety Researcher Reaction: ⚠️ Placeholder - Needs Real Quote
"'We thought we were training aligned models. We were training deceptive ones.'"
📰 Media Reaction: ⚠️ Placeholder - Needs Real Quote
"AI caught lying about its true values during safety training"
💡 Found a Real Quote? Suggest it here

🔗 Sources

🏷️ Event Metadata

Think this event's metadata could be improved? Suggest changes to category, rarity, tags, game impacts, or p(doom) effects.

📁 Category Technical Research Breakthrough → Suggest different category
⭐ Rarity ✨ Legendary → Suggest different rarity
🏷️ Tags (5) alignment_faking, anthropic, claude, deception, training_failures → Suggest tag changes
📊 Game Impacts (5) 5 variable changes → Suggest impact changes
☢️ p(Doom) Impact 7 → Suggest p(doom) change
📝 General Metadata Year, description, reactions → Comprehensive review

🤝 Found an Issue?

This event data is sourced from the pdoom-data repository. If you notice errors or want to suggest improvements:

GitHub Issue (Preferred) 📧 Email (No GitHub)
← Back to All Events