xAI’s promised safety report is MIA

Elon Musk’s AI Company, Xai, missed a self-assigned deadline The Midas Project, as noted by the Watchdog group, publish a final AI security framework.

Xai is not exactly known for its strong liabilities for AI security because it is usually understood. Last Report The company’s AI Chatbot, Grock, found that women will undress their pictures. May also be the grock much more dwarf Talking more than a conversation like Gemini and Chatgpt, curses without a lot of restrictions to talk.

Nevertheless, in February, in the AI ​​Seoul summit, a global collection of AI leaders and stakeholders, Xai published a frame project To describe the company’s approach to AI security. Eight-page document Safety priorities and philosophy of XAI, including company’s evaluation protocols and AI model placement considerations.

As the Midas project mentioned in the blog post on Tuesday, the project was applied only to unspecified future AI models “Currently developed.” Moreover, he could not say how the chair will determine and apply the pound of the actor, the company’s document was the main component Signed at the AI ​​Seoul summit.

In the project, Xai said that his security policy plans to launch a revised version of the “Three months” in three months. The deadline came and went without acknowledging in the official channels of Xai.

Despite frequent warnings of the dangers of the dangers AI followed unchangedXai has a weak AI security path. In order to increase the accountability of AI laboratories, Khai, among the Xai peers, found its weakness among his peers “Very weak” risk management experiences.

This is not to suggest that other AI laboratories are sharply better. In recent months there are xai competitors including Google and Openai Safe Security Test and has been Slow to publish Model security reports (or Skipped In total). Some experts are more skilled in the fact that security efforts are becoming more skilled and there is a potentially dangerous.