North Korea-linked Lazarus Group used Medusa ransomware in extortion attacks targeting US healthcare organisations and entities in the Middle East. The campaigns involved tools including Comebacker and Blindingcan, and Medusa has claimed hundreds of victims as a ransomware-as-a-service operation.

Multiple critical software vulnerabilities prompted remediation activity, including four SolarWinds Serv-U remote code execution flaws enabling root access on unpatched servers. A separate GitHub Codespaces issue, dubbed RoguePilot, enabled Copilot prompt injection through crafted Issues to exfiltrate GITHUB_TOKEN values and take over repositories.

Data breach activity affected major consumer platforms, with ShinyHunters claiming responsibility for intrusions at Odido and CarGurus involving millions of user records. Wynn Resorts also confirmed employee data theft following an extortion threat linked to the same group.

Artificial intelligence systems featured prominently in both offensive and defensive contexts. DeepSeek, Moonshot and MiniMax generated more than 16 million Claude queries through fraudulent accounts to distill model capabilities, while underground forums documented AI tools drafting phishing content and supporting social engineering workflows.

Supply chain and ecosystem abuse extended across developer and advertising platforms. A self-spreading npm campaign dubbed SANDWORM_MODE targeted developers with typosquatted packages, while the 1Campaign cloaking service enabled malicious Google Ads to evade detection and direct victims to phishing infrastructure.

North Korean state-backed Lazarus actors are using the Medusa ransomware in campaigns targeting organisations in the US healthcare sector and the Middle East, according to new threat intelligence. Researchers link the activity to tools previously associated with Lazarus, including the Comebacker backdoor and Blindingcan RAT, although the specific sub-group involved remains unclear. Medusa, operated as a ransomware-as-a-service since 2023, has claimed more than 366 victims, with recent attacks including US non-profits and healthcare-related entities.

Researchers have identified a new variant of the SURXRAT Android remote access trojan that conditionally downloads a large language model exceeding 23GB from Hugging Face, signalling experimentation with AI-enabled capabilities. Marketed via Telegram under a malware-as-a-service model, SURXRAT offers affiliates extensive surveillance, remote control and ransomware-style screen locking features, and shows code links to the earlier ArsinkRAT family. The malware abuses accessibility permissions and uses Firebase infrastructure for command-and-control, enabling real-time data exfiltration and device manipulation.

Orca researchers have disclosed a vulnerability in GitHub Codespaces that allows attackers to exploit GitHub Copilot through passive prompt injection embedded in a malicious issue. When a Codespace is launched from the issue, Copilot can be manipulated into checking out a crafted pull request containing a symbolic link to internal files, enabling the exfiltration of a privileged GITHUB_TOKEN via automatic JSON schema downloads. The exposed token could grant full control over the affected repository, illustrating a new AI-mediated supply chain attack path.

Varonis Threat Labs has uncovered 1Campaign, a cloaking platform designed to help threat actors run malicious Google Ads campaigns while evading detection. The service filters out security researchers and automated scanners, showing benign content to reviewers while directing real users to phishing pages or crypto drainers, supported by visitor profiling, fraud scoring and geographic targeting. Operated for more than three years via Telegram, the platform combines ad launch tools with advanced traffic filtering to prolong the lifespan of fraudulent campaigns.

Anthropic says it has uncovered large-scale distillation campaigns by DeepSeek, Moonshot and MiniMax aimed at extracting Claude’s capabilities to train competing models. The company identified more than 16 million exchanges conducted through around 24,000 fraudulent accounts, using proxy networks to evade access restrictions and harvest reasoning, coding and tool-use outputs.

Malwarebytes has identified a campaign using a counterfeit Zoom meeting page to trigger the silent download of a stealth-configured Teramind monitoring agent on Windows systems. The site simulates a glitchy video call before displaying a forced “update” prompt, which installs a preconfigured surveillance build designed to run invisibly and report to an attacker-controlled server. Because Teramind is legitimate commercial monitoring software, the activity may evade traditional antivirus detection while enabling persistent keystroke logging, screen capture and activity tracking.