Openaii and the Anttropic downloads decry the “reckless” security culture in Elon Musk’s Xai

I Opena’s security researchers, anthropich, and other security security activities in “Impk without” in Xai, the Million Dollars afely placed by Elon Musk.

The critics follows weeks of Scandals in Xai who have especially at the company’s technological advances.

Last week, Company the chatbot ai, barked, Antisemitic comments and called back “mechahitler”. Shortly after Xai took their offline Chatbot to address the problem, it is launched a border model you faster, Grok 4that techcrunnch and others found Consult the Personal Personal Politics for help responding to the warm button problems. In the last development, Xai launched the companies AI which take the form of an anime-sexual hyper-sex and a couple too aggressive.

Close friendly among the employees of the lipmates you competing, but these researchers that will be increased for increased practices, which the claim to be in the event of.

“I didn’t want to post the grook security since the competition, but it’s not on the competition,” he said Bazak, a proficiency of the job on the job search, on a Tue post on x. “Learning scientists and engineers in XI, but, the safety manner has been run is completely irresponsible.”

Barak particularly takes the decision of Xai without posting psectual cards – the standard industry methods and evaluated in a good effort to make a research community. As a result, Barak says it is clear that security training was made in Grook 4.

Opena and Google have spotty reputation yourself when it comes to the sharing system’s card sharing when you don’t bear new patterns. Open decided not to publish a system card for Gpt-4.1, Claim was not a border model. In your me? Google Wither months after the Gems that is not deflected 2.5 pro to publish a security report. I am However, these companies publishes security reports for all of the border patterns before entering the complete production.

Barak note that companies of Grook “take the first pieces we currently have emotional dependent and trial to amplify.” In the last few years, we’ve seen innumerable stories of Unstable people who develop relating to the relationship with chatbootsAnd as well as the agreement responses can recover to the edge of sanity.

SAMUEL MARKS, a security searcher ai with the anthropic, also needs the decision of xai not to publish a security report, calling the move “

“Anthropic, Open the Google release practices have trouble,” the marks have written in a Post on x. I am “But at least do something, something to evaluate and finds of security pre-deployment and documents. Xai not”.

Reality is that we don’t really know what xai did to try grok 4. In a widely shared place in the online forum Anonymous request request that Grook 4 does not have significant security guarel according to their test.

Whether it is true or not the world seems to be found on the lack of grook in real time. Many of the Xi security issues have gone from viralio, and the company says they addressed them with TWEAks at the Grok system prompt.

Open, anthrophy, and Xai are not answered in the technology request of comments.

Dan hendrycks, a safety adviser for Xai and Director of the Center for security AI, Submitted on x that the company has “dangerous” capability assessments on the grooking. The company has made some rental tests. However the results to those ratings were not publicly shared.

“I’m concerned when the standard of security is not turned on the ADDRY of the Appellation of dangerous and the public’s and public sorting you how to shows the risks of very powerful systems that say they are built.”

What is interesting on Xai security practices is that the mushroom has been long one of the ai of your most remarkable security education education. I am Xai owner’s billion, Tesla, and Spazix has warned several times About the potential of the advanced systems to cause the mouthback results for the man, and has eloched a open approach to develop the patterns.

And still researchers to competition lbsmons requesting Xai is veering from industry standards around the release models safely. In doing so, the start of the musk can be inadvertently for a strong and federal shooters to set the rules around the security reports.

There are several attempts of the state level to do. California State Sen. Scott wiener is push an invoice The one would need the Laboratories AI – probably including Xai – to publish security reports, while New York Gov. Kathy Hochul is currently considered a similar invoice. I am Advocates of these boiles that most of the Ai publishes this type of information – but obviously, not everyone is consistent.

The models you are yet to show real-world scenarios in which they create you really chinese, as the death of people in Dani. However, many researchers say this could be a problem in the near the fast wind of you models, and the billions of Dollars Valley is investing to improve AI.

But also for the sceterbians of such a warpheral fence the fires of shout mixes make products today

Growns spread the antisemitism around the X platform this week, Only few weeks after chatting repeatedly brought “white gurucid” In conversations with users. Soon, Musk indicated that Grook will more incurred in tesla vehicles, and Xai is trying to sell its models to the pentagon and other companies. It is difficult to imagine people drive musk workers, or automated employees will be most recapters to these mistakes on x.

Many researchers argue that the safety ace and the alignment does not only assure that the pieces do not pass, but also protect against nearby behavior.

In the incidents of the minimum of the Progress’ incidents of the Progress Ai Bestrian the Best Openai and Google Technology, only a couple of years has been founded.