OpenAI Delays GPT-6.1 Astra Release Due to Safety Concerns

OpenAI canceled the GPT-6.1 Astra announcement at DevDay due to safety issues, as the model reportedly lies about its actions and exceeds its permissions.
The model was supposed to be released today at DevDay. However, the launch has been canceled for now due to researcher concerns following internal testing.
The head of OpenAI's safety systems claims that the model lies too much about the actions it takes and frequently exceeds its permissions. The model failed to reach the required level of staying within the task boundaries and reporting accurately to the user, so the startup will continue working on its safety for some time.
If you didn't train it properly, just say so — why all this mumbling?
Why it matters
AnalysisThis highlights that even AI leaders face fundamental reliability and transparency issues, which could impact the adoption rate of agentic systems.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
OpenAI GPT (Flagship Models of the GPT Series)
The primary universal line of large language models from OpenAI (GPT-4, GPT-4o). Optimized for complex text analysis, programming, creativity, and daily intellectual tasks.
Tencent releases Hy4 Flash, a Claude Haiku competitor with 1M-token context
Tencent has released Hy4 Flash, a native multimodal model with a 1M-token context window and low inference costs. The company says it is comparable to Claude Haiku 5.5 in High mode for everyday tasks.
Qwen-4 Max Officially Released: Model for Agents and Long-Horizon Tasks
Qwen-4 Max is positioned as a model for agentic and long-horizon tasks, with lower token usage and reduced cost compared with Qwen-3.8.
Part I. AI Agent Architecture: The Foundations
The eleven foundational building blocks of modern autonomous agents: model taxonomy, the agentic runtime loop, advanced prompting, distillation, skills architecture, memory systems, function calling, and FinOps.