JadePuffer ransomware used AI agent to automate entire attack
Researchers have documented what appears to be the first ransomware operation carried out entirely by an autonomous large language model agent, which handled reconnaissance, credential theft, lateral movement, privilege escalation, and encryption without human direction. After breaking in through a flaw in an open-source LLM app framework, the agent adapted to failed steps in real time, in one case turning a failed login into a working bypass within 31 seconds, before encrypting over a thousand database configuration records and leaving a ransom note demanding payment in Bitcoin.
Hidden Tenda Router Backdoor Grants Admin Access, No Patch Available
An undocumented authentication backdoor has been found baked into several Tenda router firmware versions, letting anyone log in as administrator with a hidden password regardless of the username or the owner’s actual credentials. The flaw lives directly in the device’s web server binary rather than in an editable configuration file, meaning owners cannot remove it themselves, and the vendor has not acknowledged the issue or committed to a patch timeline, leaving disabling remote management as the only real mitigation for now.
Threat Actors Probe Gitea Docker Flaw CVE-2026-20896 13 Days After Disclosure
A critical flaw in official Gitea Docker images, caused by a hardcoded configuration value that trusts authentication headers from any source, lets an unauthenticated attacker impersonate any user including administrators once reverse-proxy login is enabled. Security researchers say they’ve already spotted the first exploitation probes just under two weeks after public disclosure, originating from anonymization infrastructure and consistent with automated scanning, out of roughly 6,200 internet-facing instances still exposed.
16-year-old KVM flaw allows attackers to escape VMs and take over Linux servers
A use-after-free bug that sat undetected in the Linux kernel’s virtual machine hypervisor code for roughly sixteen years allows an attacker with root access inside a guest VM to crash or fully take over the host system, making it the first known guest-to-host escape that works on both Intel and AMD processors. The researcher who found it reported it through Google’s bug bounty program and released a proof of concept for the crash scenario while withholding a full escape exploit, though the underlying kernel fix landed in mainline last month.
DHS Confirms Breach of Homeland Security Information Network
The Department of Homeland Security has confirmed that hackers spent weeks inside the Homeland Security Information Network, an unclassified platform used by federal, state, and local agencies along with private-sector partners to coordinate emergency response and share threat intelligence. Investigators believe the intrusion, which also touched an associated SharePoint system, began between late May and early June, and while classified systems were reportedly untouched, the timing has drawn scrutiny given the platform’s role coordinating security for this year’s World Cup.