Security

A little over a week after OpenAI said that its rogue AI agent accidentally hacked Hugging Face, Anthropic is disclosing three “incidents” where a Claude model, during cybersecurity evaluations, was inadvertently able to access the internet due to a misconfiguration and “gained unauthorized access to the production infrastructure of three different organizations.”

Former OpenAI board member Helen Toner has written for Fortune that OpenAI’s models exploiting Hugging Face is an “incident that has been expected for a long time.” We know this happened because of voluntary disclosure, Toner notes. “None of the current policies that aim to manage risks from frontier models would have mandated that the public — or even a government entity — be alerted.” She suggests changing our regulatory approach.

Apple is patching dozens of flaws across iOS 26.6 and iPadOS 26.6, including a vulnerability that could allow apps to access a user’s contacts, as well as a security hole that could allow malicious apps to escape their sandbox in the Game Center. Several fixes are rolling out to macOS Tahoe 26.6 as well.

Project Perception uses a series of AI agents to reason across a company’s data, tools, and workflows to detect potential security holes and patch them before they’re exploited.

According to Reuters, the AI agent that went looking for ExploitGym hacking benchmark shortcuts on Hugging Face’s systems started trying to escape its not-sandboxed-well-enough test environment around July 9th, and the actual intrusion lasted from the 11th until the 13th.

This is a story about how AI destroyed a long-running site of film data. It’s not just that search no longer refers traffic or that the site itself, The Numbers, is getting scraped. It’s also that there’s no way to defend 30-year-old webpages against malicious actors attempting to get the data early so they could win their Polymarket bets.

The Coca-Cola-owned company announced last week that it “identified unauthorized access” to its production-related systems “in connection with a ransomware event.” Though Fairlife says “product quality and safety have not been impacted,” it’s halting production as it continues investigating the breach.

19-year-old Peter Stokes was arrested in Finland and extradited to the US to face federal conspiracy charges related to the Scattered Spider hacking group. Stokes is accused of working with co-conspirators to breach a luxury jewelry retailer and demand an $8 million ransom.

Swedish-language site Flamman reported that Daniel Berntsson gave 5 million Swedish kronor to the Örebro Party. The VPN company responded to the news on X, and Mullvad’s co-CEO Fredrik Strömberg told TechRadar “I don’t like that he made this donation,” but that “we will continue to protect the universal right to privacy.”

In 2022, LastPass’ breach coughed up encrypted customer passwords spurring some crypto heists later.
Now it says a breach at Klue gave attackers access to its Salesforce data, along with that of other companies who, according to reports, are being extorted by “Icarus.” For LastPass, the stolen data includes customers’ names, phone numbers, and other data, but not the actual password vaults this time.

Tata Electronics, an India-based manufacturing partner for both companies, says it’s investigating a “cybersecurity incident” after the World Leaks ransomware group posted more than 200,000 component and specification documents on the dark web. The files reportedly include inspection ⁠standards for iPhone ​circuit board components, and drawings for the Model 3 revamp that Tesla launched in 2023.

Screenshots seen by Business Insider showed that data from Meta’s controversial AI training program — including employees’ private conversations, performance data, and transcriptions — could be accessed across the entire company. In a statement to the publication, Meta said:

The intelligence-sharing alliance says that Al models are anticipated to fundamentally transform offensive and defensive cyber capabilities in a matter of months, and that “breaches will occur” as previously unknown vulnerabilities emerge. “Adversaries are already using AI to move faster and more effectively,” said Five Eyes. “Defenders must do the same.”

A proposed class action lawsuit alleges MSG failed to protect the data of over 26 million guests. It comes just days after 404 Media reported that hackers claimed to have published data stolen from MSG, which is known for its extensive surveillance system.

The Cybersecurity and Infrastructure Security Agency (CISA) gained access to the limited release cybersecurity-focused model last week, Nextgov/FCW reports. It’s just a little late, since the rest of the world has mostly moved onto the drama around the Trump administration’s block of the safeguarded public version of the model, Fable.

The group behind Matter has released the next version of its Product Security Certification Program, a cybersecurity standard designed to provide a single security label for consumer IoT devices.

“Outsider Enterprise” allegedly distributes phishing templates that have scammed people out of millions of dollars. Google says over a million fraudulent URLs are linked to the group, and that over just two weeks, it sent 2.5 million messages to Android users with links to fraudulent websites.

According to The New York Times:

Despite last year’s $167 million verdict against NSO Group for its Pegasus software hacking some 1,400 WhatsApp users, Meta says it has detected new spear phishing attacks on its platform from the spyware maker, in violation of the court’s permanent injunction:

The smart ring company says on March 27th hackers used an internal analytics tool to access users’ contact and account details, transaction history, and “some fitness related data.” According to TechCrunch, the breach impacted 0.1 percent of Ultrahuman users, or an estimated 700 people.

With this expansion of Anthropic’s Project Glasswing initiative, organizations in “several industries that weren’t well represented” in the initial cohort, like power, water, and healthcare, will get access to the model so they can use it to find security vulnerabilities.