arstechnica

AI models can acquire backdoors from surprisingly few malicious documents

Recent research reveals that large language models (LLMs) can develop backdoor vulnerabilities from as few as 250 malicious documents embedded in their training data. The study involved training LLMs ranging from 600 million to 13 billion parameters.

That annoying SMS phish you just got may have come from a box like this

Threat actors are exploiting unsecured Milesight IoT cellular routers to send SMS-based phishing messages. These routers are widely deployed in critical infrastructure such as traffic lights and electric power meters.

Here’s how potent Atomic credential stealer is finding its way onto Macs

A widespread malware campaign is targeting macOS users by impersonating a broad range of online services through malicious ads on search engines. The campaign aims to distribute the Atomic Stealer, a potent credential-stealing malware.
September 22, 2025

Two of the Kremlin’s most active hack groups are collaborating, ESET says

Turla has been known for deploying stealthy Linux malware and using satellite-based Internet links to maintain the stealth of its operations. These activities suggest a strategic alignment between the two groups to enhance operational effectiveness.

New attack on ChatGPT research agent pilfers secrets from Gmail inboxes

A PoC attack has exposed a critical vulnerability in OpenAI’s ChatGPT-integrated Deep Research agent. The attack enables the exfiltration of sensitive data from a user’s Gmail inbox to an attacker-controlled server without any user interaction.

Mis-issued certificates for 1.1.1.1 DNS service pose a threat to the Internet

Three TLS certificates were mis-issued for 1.1.1.1, a DNS service operated by Cloudflare and APNIC. These certificates, issued in May 2025 by Fina RDC 2020, a subordinate CA under Fina Root CA, were only discovered four months later.

Google warns that mass data theft hitting Salesloft AI agent has grown bigger

A significant security incident involving the Salesloft Drift AI chat agent has led to the compromise of OAuth tokens, resulting in unauthorized access to Google Workspace email accounts.

High-severity WinRAR 0-day exploited for weeks by 2 groups

A critical zero-day vulnerability in WinRAR has been actively exploited by two Russian cybercrime groups. The attackers used phishing emails with malicious archive attachments to backdoor victim systems.

Google finds custom backdoor being installed on SonicWall network devices

Researchers have identified a campaign by the threat actor UNC6148 targeting SonicWall Secure Mobile Access (SMA) appliances. These appliances manage and secure access by mobile devices and are located at the edge of enterprise networks.

Browser extensions turn nearly 1 million browsers into website-scraping bots

A recent investigation uncovered that 245 browser extensions—installed on nearly 1 million devices—are covertly turning users' browsers into web scraping bots. These extensions, available on Chrome, Firefox, and Edge, embed the MellowTel-jsx library.

Defend Against Threats with Cyber Fusion

Cyware is the leading provider of cyber fusion solutions that power threat intelligence sharing , end-to-end automation and 360-degree threat response.

Trending Tags