Researchers fear safety disaster ahead of OpenAI’s Astra release

<figure> <img alt="" data-caption="" data-portal-copyright="Image: Bloomberg via Getty Images" data-has-syndication-rights="1" src="https://platform.theverge.com/wp-content/uploads/sites/2/2026/09/gettyimages-2287521404.jpg?quality=90&#038;strip=all&#038;crop=0,0,100,100" /> <figcaption> </figcaption> </figure> <p class="wp-block-paragraph">OpenAI is on the cusp of releasing its most powerful AI model yet, Astra, following <a href="https://www.theverge.com/ai-artificial-intelligence/987695/openai-astra-unreleased-model-cybersecurity-delay" target="_blank" rel="noreferrer noopener">weeks of delays</a> to shore up <a href="https://www.theverge.com/ai-artificial-intelligence/972380/open-ai-hugging-face-hack-ai-safety-warning" target="_blank" rel="noreferrer noopener">safety protocols</a> after its <a href="https://www.theverge.com/ai-artificial-intelligence/987566/ai-civilizations-opeai-hugging-face-hack" target="_blank" rel="noreferrer noopener">agents attacked real targets</a> during testing. As details about the model trickle out, researchers are <a href="https://x.com/RyanGreenblatt/status/2094996656186081642?s=20" target="_blank" rel="noreferrer noopener">warning</a> it "may be the single worst development for AI security/safety to date."</p> <p class="wp-block-paragraph">Shortly after OpenAI <a href="https://www.theverge.com/ai-artificial-intelligence/987695/openai-astra-unreleased-model-cybersecurity-delay" target="_blank" rel="noreferrer noopener">said</a> on Tuesday that it had delayed Astra's release to work on safety issues, <em>The Information</em> <a href="https://www.theinformation.com/articles/secret-technique-behind-openais-astra-model-sparks-security-concerns" target="_blank" rel="noreferrer noopener">reported</a> that Astra shows far less of its "thinking" than other frontier AI models, sparking concern it could be dangerously hard to monitor.</p> <p class="wp-block-paragraph">Most top AI systems today are built using a technology known as a tra …</p> <p><a href="https://www.theverge.com/ai-artificial-intelligence/988334/openai-astra-ai-monitoring-safety">Read the full story at The Verge.</a></p>