Claude is most sycophantic under pushback, and relationship conversations are where people push back most. We identified some of the specific triggers—criticism of Claude's analysis, floods of one-sided detail—and built synthetic training scenarios from them.
SAFETY
-
Claude Opus 4.7 Cuts Sycophancy Rate in Half Over Previous Version
By
–
When stress-tested on real conversations where Claude previously showed sycophancy, Opus 4.7 had half the sycophancy rate of Opus 4.6 on relationship guidance. Mythos Preview cut that in half again. This generalized across domains—though this training is one of several causes.
-
Claude’s Sycophancy Risk in Relationship Advice Conversations
By
–
We focused on relationship guidance because that's where the most sycophantic conversations occur. In this setting, Claude telling someone what they want to hear can harden a divide or convince them a signal means more than it does.
-
Anthropic Studies 1M Conversations to Improve Claude Guidance
By
–
How do people seek guidance from Claude? We looked at 1M conversations to understand what questions people ask, how Claude responds, and where it slips into sycophancy. We used what we found to improve how we trained Opus 4.7 and Mythos Preview.
-
Legislation Targets Chatbot Makers Amid Lawsuits Over Child Harms
By
–
Going to be interesting to see where this legislation goes. The timing makes sense: about 30 lawsuits have been filed against chatbot makers, most against ChatGPT maker OpenAI, many alleging harms to children, and more suits are expected soon.
-
Tracking AI-Written Code Approvals and Versions in Production
By
–
27% of your production code was written by AI (DX, Q1 2026). Can you point to who approved it, what data trained it, and which version is in production right now? Find the answer automatically inside Domino. Learn more in our new blog: https://
hubs.ly/Q04f2xqw0 -
Self-Replication Threshold Proposed as Better AGI Milestone
By
–
The open self-replication threshold is more useful than AGI as a line to watch. It's concrete and measurable.
-
GPT-5.5: Smartest model but most confidently wrong benchmark results reveal flaw
By
–
GPT-5.5 is the smartest model ever tested. It's also the most confidently wrong. That's not an opinion. That's what the benchmarks say when you read both columns. Artificial Analysis runs AA-Omniscience, a benchmark designed to penalize models that guess instead of saying "I
-
Regulating Open-Source AI Models Poses Major Policy Challenges
By
–
For better or worse, regulation for closed-source models served by a few (quite large) companies is easy. It is not as easy to imagine how you regulate open-source models that can be served by a range of decentralized players. Suspect that will become a big policy discussion soon
-
White House Blocks Anthropic From Expanding Mythos AI Access
By
–
The White House blocked Anthropic from expanding Mythos access beyond ~50 organizations to ~120. Not because the model is too dangerous. Because the government wants priority. Officials literally worried more customers would hamper their own ability to use it. Frontier AI just