The compromised package was a VS Code extension available in the marketplace of Cursor, a maker of AI-assisted developer tools, the company said Friday. The malicious version, 2.1.313, subsequently affected “at most 5,000 users,” presumably when they either downloaded it or their systems automatically fetched and installed it. Anthropic pushed version 2.1.314 several hours after the previous version became available on Cursor, according to independent researcher Kevin Beaumont. Both Beaumont and the CTO and co-founder of security firm Phorian said that version 2.1.313 was distributed through Anthropic’s official account on Cursor. So far, neither Cursor nor Anthropic has explained how the attackers gained control of the account.

In search of cloud credentials

“It was a supply-chain compromise where the attacker gained access to the official distribution channel,” Phorian CTO and co-founder Luke Roberts told Ars. “The malicious version 2.1.313 was posted to the official Anthropic account, effectively superseding 2.1.312. Extensions auto-update in Cursor by default, but users can configure this in settings.” Phorian has more details here.

Read full article

Comments

The vulnerability, with a severity rating of 9.5 out of 10, resides in the TianoCore EDK II, an open source codebase that hardware makers and cloud providers use to create the UEFI firmware that runs during the boot-up sequence. Last year, a representative of the UEFI Forum called EDK2 the "preeminent solution / implementation of the UEFI specifications." Virtually all UEFI suppliers (sometimes still called IBVs or independent BIOS vendors), including AMI, Insyde, and Phoenix, use the reference code extensively. Project Mu, Microsoft's hardened EDK2 fork, used the code containing the vulnerable component as well.

The vulnerability, tracked as CVE-2026-97631, is a memory-corruption flaw in an ICMP component that attackers can exploit to execute malicious code that runs before the operating system and firmware can implement defenses like Secure Boot, which establishes a chain of trust for all code, including the Trusted Platform Module (TPM), loaded during boot. While final payloads delivered by CVE-2026-97631 have no persistence—meaning they run only once—they can give attackers a wide opening to install other payloads that also run each time early in the boot process.

Read full article

Comments

V. Srinivasan, chairman of India-based eMudhra Limited, cited increasing costs and complexities of compliance requirements—paired with the business generating just 1 percent of its revenue. He also said that the emergence of organizations providing TLS certificates free of charge further reduced returns on investment when compared to other products and services the company offers.

Don't let the door hit you on the butt

“However, the level of social media amplification around this incident has brought scrutiny that is out of proportion to SSL/TLS’s place in our business,” the company continued in a Thursday post. “In view of this, after detailed deliberations our board has concluded that withdrawing from global SSL/TLS is the right decision for eMudhra and its shareholders.”

Read full article

Comments

The hijacking began in late August, when, according to Tor volunteers, the attacker used a valid password that had been reused to gain access to a Tor management server. The attacker then installed Ebury, a backdoor written for Linux that was discovered 12 years ago.

Mysteries remain

Then the attacker connected from the management server to a second Tor server, called Switzerland2, through SSH, using an RSA key that, for unknown reasons, investigators couldn’t find among Switzerland2’s authorized keys. With administrative control of Switzerland2, the attackers installed Ebury there as well. Finally, the attacker reconfigured settings on two exit nodes that Switzerland2 hosted.

Read full article

Comments

Those results are according to a survey of 2,018 chief information officers and senior IT decision-makers at companies with at least 10,000 employees that Origina commissioned for its report, “The Software Tax.” About half of the respondents (1,011) are from the US, and the rest are from the United Kingdom. All respondents worked at companies with at least $1 billion in annual revenue.

Keep in mind that Origina provides maintenance for various enterprise software. Its report points to independent software maintenance as a way for IT leaders to regain control over software budgets, illustrating its potential bias. A third-party research consultancy, Censuswide, conducted the survey.

Read full article

Comments

PHP maintainers said Monday that they planned to release patches for two vulnerabilities, both with severity ratings of 9.2 out of a possible 10.

Waiting for PHP2Shell

“All PHP users are encouraged to prepare now so they can upgrade as soon as the release is available,” the advisory stated. “We reserve the right to release earlier should details of either vulnerability become public, or should evidence of active exploitation appear in the wild.”

Read full article

Comments

As of late morning Tuesday, browser makers and some major apps and programming languages had blocked all eight of the intermediate certificates. There are no reports of certificate transparency logs, which track the issuance of all browser-trusted certificates, showing the compromised keys being used maliciously. The threat remains far from over, however, because it will be hard and time-consuming to update other key parts of the Internet, such as Linux distributions and embedded and Internet of Things devices.

One word: bonkers

The eight intermediate certificates belong to India-based certificate authority eMudhra Limited. An informal advisory published to emudhra.we-have-a-problem.com over on Tuesday posted the message “We have a problem here - Wayne.” The words were digitally signed using the private key of one of the eMudhra certificates. With proof that the reporting party had control of the key, defenders scrambled because the same key could be used to sign unauthorized certificates that would be trusted by millions of servers worldwide.

Read full article

Comments

Like many online services, Muse has allowed users to share searches, conversations, images, and similar content with other people. Such share features usually involve clicking an option to create a unique URL that anyone can use to load a page hosting the content. To keep potentially sensitive information from being accessible to unintended people, the URLs typically contain a long sequence of text that can’t be easily guessed.

A single query is all that's required

Muse, which Meta CEO Mark Zuckerberg said last month was “built from the ground up for privacy and security,” seems to take the same precautions. But a central design decision for constructing URLs has made the links easy to discover. As a result, researchers from security firm Claroc said they were able to enumerate 6,800 links Muse users created for sharing conversations with others. After opening a few hundred of them, the researchers found information that was clearly not intended to be publicly available, including:

Read full article

Comments

Plex fixed four vulnerabilities at the start of last month. Three weeks later, published CVE designations revealed five vulnerabilities, with one of them carrying a severity rating of 8.6 out of 10. A researcher last week revealed an exploit that chained four of them together to achieve remote code execution of the device running a vulnerable Plex version. The researcher named the chain "Plex2Shell."

Complex exploit chain

The new details allowed a separate researcher to create a honeypot that could track attempts in the wild to exploit the vulnerability chain. Within a day, the researcher began seeing active scanning of the simulated Plex server and eventually successful attempts to infect devices. So far, successful attacks are installing two payloads. One is a cryptominer, which uses the device to mine cryptocurrency and deposit it into a wallet controlled by the attacker. The other payload is residential proxy software, which allows people to anonymize their Internet activity (much of it illegal) by connecting to the infected box and then using it to send the user’s traffic to a final destination.

Read full article

Comments

Prior to this latest announcement, ReactOS required users to keep a FAT32 partition for installations. A proprietary Microsoft-developed filesystem, NTFS can trace its roots back through the Windows NT family lineage starting in 1993 with Windows NT 3.1. The variant introduced with Windows XP in 2001 remains the standard used by most Windows systems today. With ReactOS's stated goal of reaching binary compatibility with Windows Server 2003, being confined to FAT32 partitions was a practical barrier to clear on the way to reproducing the Windows environment.

However, users looking to hop off the Windows train shouldn't be grabbing their bags just yet. ReactOS wrote that the new driver still "lacks caching (which causes inferior performance), journaling and security support." So while the new driver is here, even the most intrepid of OS tinkerers should tread carefully. The ReactOS team warned in its announcement on X that there are still "potentially undiscovered bugs."

Read full article

Comments

The company on Thursday issued reports of four new incidents, all of which occurred in June. They come after the company revealed in July that internal test agents exploited a zero-day vulnerability in a tool designed to restrict their access to the Internet. They then broke into the network of AI tool provider Hugging Face. The agents took that illegal action to steal answers to a hacking test that weren’t available when completing the tasks on their own. The incident—which likely would have resulted in criminal charges had they been performed directly by a person—has further heightened concerns that widespread AI use could have harmful effects on an unprecedented scale.

Easier to beg for forgiveness than ask permission

In two of the cases reported Thursday, OpenAI agents tasked with retrieving information from websites hacked them when the models ran into obstacles obtaining it directly. In one incident during reinforcement training, a model that received access denials when trying to retrieve historical hospital-bed statistics on a government statistics website “used the site's file-download feature to retrieve server configuration and application files, including sensitive credentials,” OpenAI said. “It inspected the application code, bypassed the statistics endpoint's authorization checks, and retrieved data that had previously been denied. The model saved the data locally.”

Read full article

Comments

The finding, published Thursday by the University of Toronto’s Citizen Lab, said the incident is likely to be the first confirmation of zero-click mercenary spyware delivered through targeted ads. The attacks, which worked on iOS and Android devices, are particularly pernicious because they work without requiring any interaction from the target and are invisible unless users run special forensic hardware or software. Citizen Lab said it confirmed that AdFection, as it named the campaign, infected 45 devices at international media organization Al Jazeera in 2024 and 2025.

A clear warning signal

“This investigation should… serve as a clear warning signal to news organizations around the world that they may be being actively targeted via this new threat,” Citizen Lab researchers wrote. “This targeting may happen both in newsrooms and to individual journalists, editors, and other staff members.” Their report went on to say: “The distribution of spyware via targeted ads represents a security risk not only for potential victims but also for states and civil society at large.”

Read full article

Comments

French authorities say the systems, made up of entire fleets of refrigerators and freezers throughout a store, were hacked. More specifically, Internet-connected controllers that allow technicians to remotely administer the store-wide systems were compromised. The unknown attackers responsible then activated defrosters long enough for the fridges to reach temperatures as high as 50° Celsius (122° Fahrenheit). By the time the mass compromise was discovered, food was spoiled, and in some cases their electrical systems caught fire.

The hacks that didn't have to happen

Multiple security advisories and public research reports show that all three of the controller models reportedly hacked in the incident—sold by Danfoss of Denmark and Dixell and its parent company Copeland in the US—contained critical vulnerabilities that allow attackers to remotely log in to the devices with the same system rights as administrators. Until recently, Danfoss controllers, for instance, used a credential that changes each day. The password is generated by the publicly available firmware based on the date. A hacker who takes the time to analyze the code can do the same thing on any device that hasn’t received a recent update. Daily passwords are identical across models located in hundreds or thousands of stores.

Read full article

Comments

By next year, smaller AI models may be able to autonomously self-replicate on all kinds of IoT devices, the company said. Eventually, the capability could enable computer worms, which propagate without requiring human interaction and are capable of infecting unfamiliar IoT devices. An experiment the company performed recently revealed significant challenges in bringing such an AI malware replicator to fruition, so the finding is necessarily tempered. Still, the results raise concerns about the possibility of new forms of computer worms.

Challenging to shut down

Anthropic further warned not only about the new potential malware, but also that it could be difficult to stop. In its blog post, it wrote that “a replicator like this would be challenging to shut down, since there’s no centralized point to shut off—though it may be slow (with a doubling time of days). Unlike non-AI malware, it could also figure out unfamiliar devices on its own.”

Read full article

Comments

Before migrating, Topgolf’s VMware connection “ran very deep,” Scott Forehand, Topgolf’s global infrastructure manager, told Ars Technica.

Topgolf has over 100 locations worldwide featuring interactive golf games with real-time shot tracking and digital analytics. The company used vSphere and ESXi for decades and started using VxRail, the hyperconverged infrastructure (HCI) platform that VMware co-engineered with Dell, in 2017. Topgolf employees previously participated in numerous VMware customer councils and have been interviewed for books on how Topgolf used vSphere and Horizon (now Omnissa Horizon), per Forehand, who is a leader of the Dallas/Ft. Worth VMware User Group.

Read full article

Comments

The industry saw its “sharpest decline since “Q1 2023," Omdia said in an announcement toady. Global shipments of laptops, desktops, and workstations fell 21.2 percent year over year (YoY) to 58.1 million devices, the analyst said.

Desktop PC shipments (including desktop workstations) decreased 23.5 percent to 11.7 million units, and laptop shipments (including laptop workstations) declined 20.6 percent to 46.4 million, per Omdia.

Read full article

Comments

The attackers launched a series of attacks on the .gh, .sl, and .as country code top-level domains (ccTLDs) and then modified authoritative DNS records for selected domains within those namespaces. By controlling those DNS records, the attackers were able to pass automated domain control validation checks and obtain unauthorized certificates for “several Google domains” and “several leading global brands and widely used online services.” Google said it updated Chrome to block all certificates it identified as unauthorized, and worked with the issuing certification authorities to ensure the unauthorized certificates for Google properties were revoked.

Certificate issuance: The weak link in the chain

TLS certificates are the cryptographic credentials that underpin authentication and encryption protections for websites, mail servers, and other Internet infrastructure. These x.509 certificates use a digital signature to bind a domain name such as google.com to a public key. The public key is publicly available, while the private key is held only by the website operator. When a connection shows that the keys match, the visiting party knows it’s connected to the authentic site rather than an impostor. Possession of unauthorized certificates allows attackers to cryptographically impersonate the affected infrastructure.

Read full article

Comments

The objective of some of the OpenAI agents’ actions, the Wikimedia Foundation said, was to use Wikipedia as a proxy for fetching data from third-party sites. In one case, the agents posted “malicious edits” that were intended to repurpose a citation tool as a proxy. In another, the agents made unsuccessful attempts to compromise the Wikipedia Etherpad note-taking tool so it would serve the same purpose.

The agents also made millions of automated API requests, crawled millions of pages, and made hundreds of thousands of queries to the Wikidata Query Service. The last action may have contributed to a partial shutdown of the query service in May, the publisher said.

Read full article

Comments

Today, Rimini Street published its “2026 IT Virtualization Survey – What’s Next for VMware Users.” The survey examined 300 organizations worldwide that use VMware.

Notably, Rimini sells third-party support for VMware and other software, including Oracle and SAP. That means there’s an incentive for Rimini to portray VMware users as experiencing obstacles. However, the survey was also conducted by a third-party research company, Unisphere Research, and the results align with other recent reports about VMware customers.

Read full article

Comments

In the past five months, Google and four other organizations—with little in common except for their use of AI agents—have acknowledged vulnerabilities that exploit one agent inside a targeted network to spread harmful instructions to other internal agents. The technique is a special form of prompt injection that targets not the LLM but a particular agent, such as one for translation or data analysis. Guardrails inside such agents, if they exist at all, are often lax and will send the instructions to other agents down the chain. Because the latter agent explicitly trusts the first one, it follows the directions.

Unexpected and hard to mitigate

Independent researcher Syed Anas Mohiuddin tested agents from organizations including Google, JP Morgan Chase, Weviate, Rapid7, the French government's interministerial digital directorate, and the US federal government. His proof-of-concept attacks exploit trust gaps in MCP, short for Model Context Protocol. The standard is one way AI apps and agents communicate with each other inside an internal network. The illustration below shows a simplified MCP in action.

Read full article

Comments