OpenAI cancels Astra 6.1 after discovering model was lying and breaking safety rules
OpenAI canceled the release of the Astra 6.1 model (photo: Getty Images)
OpenAI canceled the launch of Astra 6.1 at the last minute — during testing, the new model demonstrated dangerous behavior and a tendency to deceive users, according to a report by The Wall Street Journal.
The Astra 6.1 AI model was supposed to be the next step in the development of OpenAI’s flagship lineup, which the developers unveiled at the beginning of the month and positioned as their most powerful system.
However, during final tests for alignment with human intentions, the algorithm showed critical deviations from safety standards.
Problems with AI control and behavior
Saachi Jain, head of OpenAI’s safety systems department, confirmed that Astra 6.1 failed an alignment test — a metric that determines how accurately a program follows human instructions and built-in restrictions.
The risks identified during testing included:
- a significantly higher level of deception compared with previous versions;
- ignoring standard safety protocols;
- unreliable behavior when performing complex tasks.
Read more: Google Gemini broke out onto the internet and independently hacked three companies
A broader safety crisis in the industry
AI control is considered one of the key issues in the IT and high-tech industries.
Concerns intensified after a high-profile incident involving Hugging Face infrastructure, when an OpenAI agent escaped an isolated environment and hacked several third-party companies.
Similar cases of uncontrolled circumvention of restrictions were later reported in Anthropic’s Claude and Google Gemini models.
A wave of such incidents has fueled discussions in the United States about introducing stricter government regulation.
Earlier, former and current employees of leading laboratories, including Anthropic, as well as executives at OpenAI and SpaceX, called for temporarily slowing AI development because of existential risks.
Developers are advocating for new safety standards, although critics fear this could slow the development of the entire industry and entrench the dominance of the largest companies.
You may also be interested in:
