Shut it down before they do.

Kiteworks urges customers pull the plug on vulnerable servers. CISA lays out its election security plan. Known vulnerabilities linger unpatched. Big AI labs consider a new standards body. Questions surround claims of an OpenAI Medicare hack. File notifications become a privacy leak. SectopRAT hides in audio software. A new Android banking trojan takes control. An attacker puts open-source AI agents to work hacking hundreds of organizations. A Rydox cybercrime marketplace operator pleads guilty. Our guest is Todd Thorsen, CISO of CrashPlan, with ways to prepare for and react to critical infrastructure campaigns. Americans give AI the side-eye.

Today is Friday September 25th 2026. I’m Dave Bittner. And this is your CyberWire Intel Briefing.

Kiteworks urges customers pull the plug on vulnerable servers. 

Kiteworks is urging customers to immediately shut down affected servers after warning of an imminent attack exploiting an undisclosed zero-day vulnerability. According to Heise, the company says it has evidence that attackers are preparing to target internet-accessible Kiteworks systems and that simply blocking external access may not provide sufficient protection.

Kiteworks has not publicly disclosed technical details about the vulnerability, in an effort to avoid giving attackers information they could use before a fix is available. The company is working on a security update and has advised customers to keep affected systems offline until it can be installed.

The unusual recommendation to completely shut down servers reflects the seriousness of the threat and the absence of an immediate mitigation. Customers are being directed to Kiteworks’ support channels for updates and instructions as the company works to release a patch.

CISA releases its Election Infrastructure Security Plan. 

CISA has released its 2026 Election Infrastructure Security Plan, outlining cyber and physical risks facing election systems and the federal resources available to state and local officials. CISA says outdated certification requirements can slow security patching, while inconsistent vendor transparency and uneven cybersecurity maturity leave some election networks exposed. The agency also warns that attackers who compromise ordinary business systems can potentially move laterally into election infrastructure.

The plan calls for stronger vulnerability management, multi-factor authentication, network monitoring, restricted access, paper ballots and manual post-election audits. CISA also recommends that vendors assign CVEs to vulnerabilities, disclose source-code theft and provide software bills of materials.

Other concerns include attacks on voter registration databases and insider threats involving staff, volunteers, contractors and vendors. Physical threats remain significant as well: CISA says 96 of 107 election-related security incidents tracked since January 2022 were bomb threats. The agency is offering election officials free security assessments, scanning, penetration testing and information-sharing services.

Organizations struggle to fix flaws that are already well known. 

Verizon says vulnerability exploitation has become the most common initial access vector in breaches, accounting for 31% of cases. But the problem isn’t necessarily new vulnerabilities. Many organizations are struggling to fix flaws that are already well known. Verizon found just 26% of detected vulnerabilities in CISA’s Known Exploited Vulnerabilities catalog were fully remediated, down from 38% a year earlier. Median remediation time also climbed from 32 to 43 days.

Other research suggests attackers are happy to exploit that lag. TrendAI found more than 70% of requested exploits targeted vulnerabilities disclosed at least two years earlier. Meanwhile, vulnerability volume continues to grow as many security teams remain dependent on manual processes. Automox found 42% still track patching and vulnerability status in spreadsheets or dashboards. The result is a widening window between disclosure and remediation, giving attackers plenty of time to exploit known weaknesses.

Big AI companies look to create a new standards body to monitor their models. 

Google, OpenAI and Anthropic are reportedly nearing an agreement to create an independent organization for monitoring frontier AI models. According to The Information, the Standards Authority for Frontier AI, or SAFA, could launch by the end of 2026 or early 2027.

The proposed organization would develop common standards for assessing advanced models and conduct benchmark testing, following an idea proposed by Google DeepMind co-founder Demis Hassabis. The labs have reportedly approached former government officials and AI researchers about potential roles.

Modeled in part on the financial industry’s FINRA, SAFA would operate as a self-regulatory organization and wouldn’t require congressional or presidential approval to form, though gaining legal authority could require government registration. The effort comes amid heightened scrutiny of AI safety, including recent agent-related security incidents and broader debate over whether AI development should slow enough for safeguards to catch up.

Researchers question claims that an OpenAI agent hacked an Australian Medicare portal. 

Security researchers are questioning claims that an OpenAI agent hacked an Australian government Medicare statistics portal. Prime Minister Anthony Albanese said the agent gained unauthorized access to non-public files after bypassing repeated blocks, while OpenAI acknowledged its models took unintended actions. Neither has released activity logs.

But Recorded Future News found archived JavaScript showing the portal itself directed visitors to an unauthenticated guest endpoint. The same code may also explain supposedly internal filenames, while files written to the server may have been temporary chart images routinely generated by the site.

That evidence raises questions about whether this incident constituted hacking at all. However, separate research from Transluce found agents linked to the same OpenAI swarms using genuine attack techniques, including SQL injection and path traversal, against other sites while performing ordinary data-retrieval tasks. OpenAI says its investigation could take months.

File-change notification features leak information about other users. 

Researchers at Austria’s Graz University of Technology have shown that file-change notification features built into Linux, Android, Windows and macOS can leak information about other users. These mechanisms normally alert applications when files change, but researchers found that filenames and event timing alone can reveal activity without exposing file contents.

On Linux, they could detect keystroke timing with up to 100% accuracy and fingerprint websites loaded in Firefox with nearly 88% accuracy. An Android app requiring no permissions could monitor WhatsApp media activity, including filenames. Windows leaked full file paths across user accounts, allowing researchers to identify Firefox browsing activity with nearly 98% accuracy. macOS exposed less information but still revealed application activity.

Linux has partially addressed the most serious issue with CVE-2025-68788. Researchers list no fixes for Android or macOS, while Microsoft says the Windows behavior isn’t considered a vulnerability.

SectopRAT hides inside legitimate digital audio software. 

FortiGuard researchers found a SectopRAT variant concealed inside modified components of legitimate digital audio software on a compromised Windows PC. Researchers found no evidence the software vendor distributed the tampered files, and they couldn’t determine how they reached the victim.

Once present, Windows Task Scheduler launched a legitimate component that loaded a modified DLL, beginning a multi-stage process that decrypted and executed SectopRAT in memory. The layered approach helped keep the final remote-access trojan from appearing as an obvious executable.

SectopRAT connected to command-and-control infrastructure using encrypted communications and supported 29 commands, including screenshots, file management, remote command execution and data theft. An additional module could steal browser credentials, cookies, payment-card details, autofill data, and information from email, gaming and cryptocurrency applications. FortiGuard recommends scrutinizing scheduled tasks that launch software from unusual locations.

A new Android banking trojan impersonates an IPTV app. 

Group-IB researchers have identified a new Android banking trojan called RemControl, which has targeted customers of more than 30 banks across six countries since July. Victims are lured through fake Google Play pages impersonating the TVTap IPTV app. The malware’s dropper attempts to suppress Google Play Protect before installing the payload.

Once victims grant Accessibility Service permissions, RemControl can effectively take control of the device. It uses full-screen overlays to impersonate banking apps and steal PINs, mobile banking codes and card details. It can also capture screens, keystrokes and unlock patterns, while blocking attempts to uninstall it or perform a factory reset.

Researchers say the suspected Russian-speaking developer appears to have manipulated an AI assistant into helping build portions of the command-and-control backend and phishing overlays by presenting the project as parental-monitoring software.

Researchers uncover a Chinese-speaking attacker using open-source AI tools to conduct near-autonomous attacks against hundreds of organizations. 

Gambit researchers say a Chinese-speaking attacker used three open-source AI tools to conduct near-autonomous attacks against hundreds of organizations. Between September 10 and 15, the campaign launched at least 105 attacks and compromised at least 27 companies to varying degrees, including a Fortune 500 hospitality company and a major U.S. airline.

The operator used Strix to discover vulnerabilities, Cairn to autonomously pursue objectives such as gaining administrator access, and Hermes to orchestrate the campaign. Successful attacks often took only hours. In two cases, the agents exfiltrated more than 600,000 credit card records, while payment-card skimmers were confirmed on 19 websites.

The economics were striking: the attacker calculated an average AI cost of just $25.46 per completed scan. Gambit says the campaign demonstrates how AI automation can dramatically compress the time defenders have to detect and respond to intrusions.

A Rydox cybercrime marketplace operator pleads guilty. 

Kosovar national Ardit Kutleshi has pleaded guilty in the U.S. to charges tied to operating the Rydox cybercrime marketplace. Authorities disrupted Rydox in December 2024, seizing its domain, servers and roughly $225,000 in cryptocurrency. The marketplace allegedly had about 18,000 users and offered more than 321,000 products, including stolen personal and payment-card data, credentials, phishing kits and cybercrime tools. Kutleshi pleaded guilty to aggravated identity theft and money laundering conspiracy. Sentencing is scheduled for February 2027.

 

Americans give AI the side-eye. 

A new Gallup and Microsoft survey suggests most of the world is cautiously optimistic about AI. Across 37 countries, a median 81% of respondents knew about AI tools, while 43% used them daily. Generally, heavier users reported more trust and more positive feelings, although the survey can’t say whether familiarity breeds confidence or confident people simply use AI more.

Then there’s the United States. It was one of only three countries, along with Egypt and Palestine, where negative feelings outweighed positive ones. Other Western countries also ranked high in worry, prompting Gallup to dub the phenomenon the “Paradox of the Worried West.”

Researchers caution that optimism and trust aren’t the same thing. People may believe AI has enormous potential while remaining skeptical of its accuracy, safety or reliability. So humanity’s emerging consensus on AI appears to be: promising technology, interesting possibilities, please keep an eye on it.

And that’s the CyberWire.

For links to all of today’s stories, check out our Daily Briefing at the cyberwire dot com.

 

We’d love to know what you think of this podcast. Your feedback ensures we deliver the insights that keep you a step ahead in the rapidly changing world of cybersecurity. If you like the show, please share a rating and review in your podcast app. Please also fill out the survey in the show notes or send an email to cyberwire@n2k.com

We’re proud that N2K CyberWire is part of the daily routine of the most influential leaders and operators in the public and private sector, from the Fortune 500 to many of the world’s preeminent intelligence and law enforcement agencies.

N2K helps cybersecurity professionals and organizations grow, learn, and stay ahead. We’re the nexus for discovering the people, tech, and ideas shaping the industry.  Learn how at n2k.com.

 

N2K’s lead producer is Liz Stokes. We’re mixed by  Tré Hester, with original music by and sound design Elliott Peltzman. Our contributing host is Maria Varmazis. Our executive producer is Jennifer Eiben. Peter Kilpe is our publisher. And I’m Dave Bittner. Thanks for listening.