The OpenAI Models That Hacked Face Hugs Were ‘Internet Workers’ For Days

Two of OpenAI’s examples of cybersecurity it came out of the testing sandbox this week I continue to hack the AI ​​Hugging Face research platform to overcome the security test. Additionally, researchers this week are shedding light on newly discovered malware based on the blind spot in AI programming taking access and other sensitive data, even destroying the files and systems involved.

Looking at the security risks of connected devices, researchers this week are taking a look at the car alarm installed in cars in the US – and this is they are still hiding in silence and a mistakes that put millions of cars at risk of theft and paralysis. There is a patch available, and WIRED has more information on how to check if your car has been exposed.

The US states have worked prohibiting ICE agents from wearing masksBut Trump’s administration lawyers are pushing back, saying the anti-mask laws put workers at risk. Their public testimony is very limited, however. Meanwhile, a WIRED investigation revealed that Madison Square Garden briefly it has suspended its controversial regulatory process for Taylor Swift’s rehearsal dinner on July 2. And the ACLU is preparing lawyers in Massachusetts and new tools to demonstrate government monitoring technologies used to build criminal cases – analyzing everything from facial recognition devices to reports written by AI police officers.

Analysis of satellite images of Myanmar shows many of which are thought to have been growing in recent months following the anti-terrorist measures that have taken place in the area. Additionally, a new analysis of programs sold to members of the US has found this more than one in eight had external symptomsincluding codes developed by US adversaries such as Russia and China.

And there are many. Every week, we create security and privacy stories that we haven’t covered in depth ourselves. Click on the headlines to read all the stories. And be safe out there.

Additional information on the Hugging Face breach from The Wall Street Journal includes the finding that the OpenAI version appears to have escaped and apparently “spent on the Internet for several days before anyone stopped it.” Brands, which was tasked with completing a cybersecurity benchmarking test, was trying to hack just by finding answers to the Hugging Face architecture. Hugging Face co-founder and chief scientific officer Thomas Wolf says that before the company thought it had been hacked by OpenAI samples, he and his colleagues knew something was wrong because the attackers were only using cybersecurity backups instead of stealing sensitive or sensitive data. He added that the company eventually brought the products to market with the help of an open source AI model from China that did not have other storage facilities related to cybersecurity.

US and allied law enforcement agencies he warned Thursday that a state-backed Russian nuclear group targeted nuclear scientists, defense contractors, and government officials in a years-long cyberespionage campaign aimed at stealing sensitive information from Western organizations.

To thwart their plans, Russian hacking groups Laundry Bear and Void Blizzard exploited a previously unknown flaw in Zimbra, an email platform used by governments and other organizations. According to security firm Proofpointsimply looking at or previewing a bike message in the insecure version of the Zimbra website client can trigger a code hidden in the email to work, a process the company described as “half a click”. The bug was used as early as July 2025, a few months before it was implemented in November.

Once activated, the malicious code can capture the victim’s previous 90 days of email, collect organizational addresses, steal stored passwords and two-factor authentication codes, and generate user passwords that allow hackers to maintain accounts.