Latest cybersecurity reporting from selected sources.

Unauthenticated Flaw Allows Full Router, Network TakeoverA hidden backdoor, disclosed by CERT/CC and found in multiple firmware versions made by Chinese manufacturer Tenda, bypasses authentication and could grant attackers administrative access. Researchers are reporting exploitation and an Nmap script is making vulnerable devices easier to identify.

Unauthenticated Flaw Allows Full Router, Network TakeoverA hidden backdoor, disclosed by CERT/CC and found in multiple firmware versions made by Chinese manufacturer Tenda, bypasses authentication and could grant attackers administrative access. Researchers are reporting exploitation and an Nmap script is making vulnerable devices easier to identify.

Eset Links Group's Growth to Integrated Endpoint-Killing ToolsEset researchers say the rapidly growing Gentlemen ransomware operation differentiates itself by supplying affiliates with a standardized EDR-killer suite that disables security tools, quickly incorporates newly disclosed vulnerable drivers and helps scale attacks across multiple regions worldwide.

Claroty Warns of Downtime, 'Devastating' Impact of Vulnerabilities in OT SystemsVulnerabilities in backup power devices and heating and cooling control systems widely used in data centers could enable remote cyberattacks and result in costly downtime and "devastating" operational impact, according to new research from OT security firm Claroty.

Security Researchers Fear Broader Legal Pressure on Bug DisclosuresMicrosoft is pursuing legal action after a researcher publicly released six Windows zero-days and exploit code following a breakdown in coordinated disclosure talks, escalating tensions over vulnerability disclosure, platform moderation and protections for independent security researchers.

Expect to See Widespread Availability of Mythos-Level Models Within 6-12 MonthsAnthropic is expanding public access to its frontier artificial intelligence model Claude Mythos "to qualifying customers' security teams on request" for such purposes as vulnerability research and red-teaming, and predicts that Mythos-class models will be publicly available within 12 months.

Study Finds Standard Safety Tests Miss Most Agentic AI ThreatsResearchers from Stanford, MIT, Carnegie Mellon and others found that most production AI agents are vulnerable to attacks that unfold across multi-step actions. The study warns that memory, tool access and agent coordination create failure modes traditional chatbot safety testing cannot see.

No Patches Yet Available, After Third Party Published Vulnerability DetailsSecurity researchers have discovered a new, critical flaw in the Linux kernel that attackers can exploit to gain root access. No patches are yet available to fix "Dirty Frag," the second new local privilege escalation flaw to be found in two weeks, following the similar "Copy Fail" vulnerability.

AI Tool Used to Discover Bugs, Which Included 2 Maximum Severity VulnerabilitiesResearchers at security firm AISLE said they recently identified 38 vulnerabilities, including two maximum-severity zero-day flaws in OpenEMR, an open-source electronic medical record software platform used by about 100,000 healthcare providers globally. OpenEMR has patched the problems.

UPGRADE and DigiSeals Programs at ARPA-H Remain Fully FundedA U.S. federal grant effort to develop autonomous medical device patching platforms for hospitals evaded the budget-cutting knife of the Trump administration. Program boosters hope to automate cyber defenses so that hospitals of any size can more quickly patch vulnerabilities.

AWS Researchers Find an Interlock Server Laden With ToolsRansomware hackers exploited a flaw with a maximum vulnerability score in Cisco firewall management software weeks before the networking giant disclosed the vulnerability in early March. The group has focused extensively on critical infrastructure sectors in North America and Europe.

Flaws Let Attackers Run Commands and Steal API Keys Before Trust PromptCheck Point research found three critical flaws in Anthropic's Claude Code that allow attackers to execute arbitrary commands and steal API keys through repository configuration files, before users see a trust prompt. The AI giant has patched all three vulnerabilities.

LLM-Built Toolkit Hit 91 Hosts, Mined Funds in MoneroSecurity researchers detected artificial intelligence-generated malware exploiting the React2Shell vulnerability, allowing attackers with no coding expertise to build functional exploits. The attacker may have circumvented an AI model's safeguards by framing the malicious coding request as homework.

AI Assistant Executes Hidden Commands Embedded in Docker Image LabelsA vulnerability in Docker's Ask Gordon AI assistant allows attackers to execute malicious commands by hiding them in the container application development platform's image metadata, said security researchers. Dubbed DockerDash, the vulnerability exploits a failure across Docker's AI execution chain.

Chinese Attackers Among Those Tied to Attempted Exploits of FortiSIEM AppliancesCritical vulnerabilities in edge devices are continuing to be discovered by security researchers and rapidly targeted by attackers. Lately, this includes a critical vulnerability in Fortinet's FortiSIEM appliances, which Chinese and other hackers began targeting just two days post-patch.

AI Agent Can Access File Upload API to Exfiltrate DocumentsSecurity researchers have demonstrated how Anthropic's new Claude Cowork productivity agent can be tricked into stealing user files and uploading them to an attacker's account, exploiting a vulnerability the company allegedly knew about.

Israeli Startup Novee's Custom AI Agents Mimic Human Attackers to Scale DetectionNovee launched with $51.5 million in funding to build AI agents trained to find and fix vulnerabilities. Its proprietary model combines human-led research, prompt engineering and simulations to offer scalable penetration testing as AI-fueled threats outpace traditional tools.

3 Major Tech Firms Shipped Vulnerable Open-Source Tools to Hugging FaceResearchers discovered remote code execution vulnerabilities in three AI libraries from Apple, Salesforce and Nvidia used by models with tens of millions of Hugging Face downloads, allowing attackers to hide malicious code in model metadata.

Patches Issued for MongoBleed as Ransomware Groups Target Flaw to Steal DataTens of thousands of internet-exposed MongoDB databases are at risk as attackers actively target a critical vulnerability in the software to steal sensitive data, with ransomware groups having joined the fray, researchers warn. MongoDB has issued patches and mitigation advice.

Unpatched Flaw in Open-Source Gogs Service Facilitates Remote Code ExecutionAn attacker has been exploiting a zero-day vulnerability in Gogs, an open-source and popular Git service that allows for self-hosting, warned researchers. At least 700 internet-exposed servers running Gogs shows signs of being infected with command-and-control malware; no patch is yet available.