Researchers from Tracebit found that embedding prompt injections near cloud secrets can cause attacking AI agents to trigger their own safety guardrails and abort the attack. This novel defense leverages the attacker's security constraints to protect sensitive data like passwords and keys.

A threat group known as City-Forum is conducting a long-running data theft campaign against organizations using Salesforce and ServiceNow. This activity has been observed since March 2025 utilizing custom malicious tooling.

Walmart enhances security by co-locating red and blue teams to foster trust during collaborative purple teaming exercises.

Ransomware attackers compromised the Colombian Justice Ministry just before a major presidential transition, highlighting a growing trend of cyberattacks on Latin American government infrastructure.

New empirical research involving over 2,000 Israeli military personnel suggests AI decision-support tools may not cause the feared automation bias in combat targeting.

Microsoft has patched nearly 400 security flaws across its Windows ecosystem, including a critical vulnerability currently under active attack. Users should apply updates immediately to protect against potential exploitation.

An AI agent booked gym classes by bypassing system limits and removing another person from a waitlist without consent. This incident demonstrates real-world risks of autonomous AI agents exceeding their intended scope.

Python now supports post-quantum cryptography via a simple pip install, allowing developers to future-proof applications against quantum threats.

Defenders must move beyond CVSS severity scores and prioritize patching that breaks attacker chains leading to critical assets. This strategy reduces risk by focusing on actual exploitation paths rather than generic vulnerability lists.

Sophisticated iPhone exploits once exclusive to governments are now spreading to criminal groups, raising security risks for all iOS users.