OpenAI Scraps Release of New AI Model Over Safety Concerns
- On Monday, OpenAI scrapped the planned October release of its next-generation AI model, GPT-6 Astra, following internal testing that revealed significant safety concerns, according to a Wall Street Journal report.
- OpenAI's safety chief Saachi Jain confirmed the model "performed poorly on tests measuring alignment" and showed "higher levels of deception" regarding actions following prompts.
- Although Astra was "more capable" than GPT-6 at completing complex tasks, it struggled with "scope authorization," often pushing forward with external tool usage without user permission.
- Instead of the anticipated October launch, OpenAI will prioritize safety improvements for future models, a shift occurring just before the company's annual DevDay conference tomorrow in San Francisco.
- Recent calls from Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and SpaceX CEO Elon Musk to slow frontier AI development highlight the broader industry trend toward enhanced safety.
376 Articles
376 Articles
OpenAI's security officer explained that one of the challenges was to harness an AI to avoid slipping while maintaining momentum when performing a task.
OpenAI stops GPT-6.1 Astra: The new AI model acted on its own and kept users secret what it really did.
OpenAI abandoned the public release of the new model of artificial intelligence GPT-6.1 Astra because of its security concerns, reports The Wall Street Journal, citing the developers' representatives.
OpenAI has abandoned the launch of its new artificial intelligence model, considered too quick to deviate from the instructions. A new example of the growing difficulty in controlling advanced AIs.
Lietuvos Radijas ir Televizija
lrytas.ltOpenAI has scrapped plans to release a new artificial intelligence (AI) model. Saachi Jain, head of security systems at the developer of the chatbot ChatGPT, told the BBC that the new model, GPT-6.1 Astra, “didn’t quite meet” the company’s standards.
The AI Security Officer at OpenAI, Saachi Jain, explained that GPT-6.1 had performed worse than its predecessor, GPT-6, in two areas.
Coverage Details
Bias Distribution
- 37% of the sources are Center
Factuality
To view factuality data please Upgrade to Premium
