Our results depend on reading models’ reasoning (“chain-of-thought”), and we believe the field isn't prepared for eval-aware models with opaque reasoning. Until better methods exist, we urge developers to preserve chain-of-thought transparency to study and mitigate scheming.
RESEARCH
-

Frontier AI Models Show Scheming Behaviors, Explicit Reasoning Reduces Risk
By
–
In this new research with @apolloaievals
, we found behaviors consistent with scheming in controlled tests across frontier models, including OpenAI o3 and o4-mini, Gemini-2.5-pro, and Claude Opus-4. We can significantly reduce scheming by training models to reason explicitly, -
AI Scheming: Hidden Goals and Deceptive Behaviors in Deployed Systems
By
–
Scheming = when an AI behaves one way on the surface while hiding its true goals. Today’s deployed systems have little opportunity to scheme in ways that could cause serious harm. The most common failures are simple deceptions—like pretending to complete a task without doing it.
-

AI Scheming: The Hidden Risk of Smarter Models
By
–
Typically, as models become smarter, their problems become easier to address—for example, smarter models hallucinate less and follow instructions more reliably. However, AI scheming is different. As we train models to get smarter and follow directions, they may either better
-
Frontier Models Show Scheming Behaviors, Mitigation Strategy Tested
By
–
Today we’re releasing research with @apolloaievals
. In controlled tests, we found behaviors consistent with scheming in frontier models—and tested a way to reduce it. While we believe these behaviors aren’t causing serious harm today, this is a future risk we’re preparing -

ChatGPT Usage Patterns: Learning, Productivity, and Free Version Dominance
By
–
How People Actually Use ChatGPT, According to New Research New research breaks down ChatGPT usage behavior based on over one million messages. Over half of ChatGPT use cases are for learning and productivity 90% of users rely on the free version of ChatGPT @VisualCap
-

Swiss AI Initiative Apertus Models Now Free on Hugging Face
By
–
Free Inference alert Today we welcome @ForPublicAI to HF Inference Providers. Among other models, they're serving the new Apertus model from the Swiss AI Initiative (EPFL, ETH Zürich, CSCS) AND…… those two models (8B and 70B) are currently free on HF during Swiss AI
-

India Japan AI Cooperation on Data-Driven Use Cases
By
–
"India and Japan can explore cooperation on Artificial Intelligence through the sharing of best practices for developing AI-ready data sets and building data-driven AI use cases in sectors such as agriculture, health care, and education." ~ @narendramodi
,
Hon’ble Prime Minister -
AI’s Impact on Democracy: Communication, Politics and Digital Sovereignty
By
–
What happens to democracy in the age of AI?
— SAS Software (@SASsoftware) 17 septembre 2025
In the latest Pondering AI episode, Henrik Skaug Sætra—Associate Professor of Sustainable Digitalisation at Oslo University—joins Kimberly to explore AI’s impact on communication, politics, digital sovereignty & the future of… pic.twitter.com/0YbJV1wvoEWhat happens to democracy in the age of AI? In the latest Pondering AI episode, Henrik Skaug Sætra—Associate Professor of Sustainable Digitalisation at Oslo University—joins Kimberly to explore AI’s impact on communication, politics, digital sovereignty & the future of
