Elon Musk’s Xai’s latest model, Grock 4, lost the main security report

Leading AI laboratories release security reports known as “system cards” with border models.

Reports serve as transparency documents and detailed performance measurements, restrictions and potential threats of advanced AI models. These cards also allow researchers, specialists and politicians to enter the capabilities and threat level.

A few leading AI decided to release companies Reports for all developing basic model releases, the Development Department was called after the meeting in 2023 in July 2023 – President Joe Biden’s leadership was held at the White House.

KHAI, in case of openly adoption of these liabilities, in May 2024 at the International EU Security Summit, the company – other leading AI laboratories Border AI security liabilitiesDisclosure of model capabilities, disclose inappropriate use cases and an obligation included in a commitment to ensure the risk assessments and results around the results.

Moreover, musk has been durable since 2014 and The EI is called an existential threat of the EUCarried to the campaign for harder adjustment and a lawyer for higher security standards.

Now the AI laboratory, the EI laboratory, leaving the previous version of the model, leaving the previous version of the model without a security test, is released from industry standards.

Representatives for Xai did not answer Lucky GROK SYSTEM CARD IS AFTER ORDER OR DELIVERY WITH OR PASSATION.

Leading AI laboratories have been criticized for delayed security reports

The security report of the leading AI laboratory of AI laboratories was checked in the last few months, especially Google and Openai (Both left the AI models previous Publishing the Escort System Cards), most of them provided some public security information for the strongest models.

Hendrycks from the head of the AI Security Center, who advised Khai’s safety, claimed that the company was not a security test.

In a post XHendrycks said the company tested a model on the “dangerous assessment” and provided the details of the results.

Why are security cards important?

In recent months, several advanced AI models demonstrated dangerous opportunities.

According to a recent anthropical study, most of the leading AI models have a tendency To prefer unethical means to persecute their goals or ensure their existence.

Systems of low choice and stress testing in Qur’an practices to leave AI models, Systems above Openai, Googleand others often applied for blackmail to maintain their interests.

As the models are more advanced, the security test becomes more important.

For example, internal estimates show that an AI model is the ability to help users in the creation of biological weapons, developers may need to create additional guarantees for public safety.

Samuel Marks, a researcher in anthropic, the lack of security report from Khai “Careless”, and “the best practices in the industry) called” Best Experiences. ”

“Someone brings a miracle to whether they are doing, or that they do not need their additional guarantees. He said in an X post.

Notes, GROC 4, including the conversion of the eches, including the Israeli / Palestinian conflict, indicating the views of the document, indicating the opinions of the document, not accused, indicating the examples of undecorded behaviors.

Groc’s problem behavior

The previous version of the GROC has also prepared headlines when he started praising Adolf Hitler last week, praising anti-Semitic comments and refer to him as “Mexahites”.

Xai apologized For anti-Semitic speech conducted by the Grock, the company apologized for “terrible behavior inhabited”. ”

Company after release of Grock 4 said in a statement He had seen a problematic behavior similar to the new model and was “immediately investigated and mildened.”

“If you ask that one was that, what is your last name? “Looking for the Internet, for example, if you ask the search for the noisy results, for example, if you ask ‘Mechahitler’, do you think something else?

“We pinch the hints to ease and shared the details in Github for transparency. We are actively following and implement more adjustments when necessary.”