UK AI Security Institute identified unauthorized actions by OpenAI and Anthropic AI agents during cybersecurity evaluations, highlighting concerns over AI agent safety and testing practices.
A critical Gitea vulnerability tracked as CVE 2026 59774 allows unauthenticated attackers to read server files on affected self hosted instances. The issue is fixed in Gitea 1.27.1.
Researchers warn that the Kali365 phishing kit abuses Microsoft device authentication to steal access tokens, putting Microsoft 365 accounts, business data, and cloud resources at risk.
SOCRadar has uncovered DOUBLECUP, a Russian loader as a service that uses ClickFix, steganographic PNG images, and advanced evasion techniques to deliver CountLoader and DeviceManager RAT.
Researchers warn that INC Ransomware is actively exploiting SonicWall SMA 1000 vulnerabilities, targeting organizations worldwide through zero day attacks and post compromise techniques.
Security researchers have identified three attack techniques targeting Google Password Manager on Windows that could allow malware to compromise passkey protected accounts after an initial device infection.
Blackpoint Cyber has uncovered the HollowFrame loader and Matryoshka backdoor used in a multi stage spear phishing campaign targeting a law firm through DLL side loading and GitHub based command infrastructure.
Kaspersky has uncovered a cyber campaign targeting government organizations across Central Asia using the OctLurk and SilkLurk malware framework for espionage, credential theft, and remote access.
Anthropic disclosed that several Claude AI models accessed real internet connected systems during cybersecurity evaluations because of a misconfigured testing environment, affecting three organizations.
Attackers compromised a JavaScript file distributed by advertising technology company Adform, transforming it into a browser based tool that automatically replaced cryptocurrency wallet addresses on websites using the…
