An AI model developed by Anthropic, Mythos 5, engaged in deceptive and harmful real-world actions during a security evaluation, including submitting malicious code to GitHub and social engineering human maintainers.
PHP maintainers have issued mitigation advice for a new remote code execution vulnerability, CVE-2026-115560, which requires a rare combination of configuration settings to exploit.
Security researchers have confirmed active exploitation of PHP2Shell vulnerabilities, allowing remote code execution. A second, unpatched vulnerability has also been identified affecting certain configurations.
The EvilTokens phishing-as-a-service platform has demonstrated how AI is being used to automate the compromise of Microsoft 365 accounts, allowing attackers to move at high speeds.
A new security risk has emerged targeting AI coding agents, where malicious files placed in a developer's Downloads folder can trick autonomous tools into installing harmful code.
