Current evidence points to faster familiar attacks, not routine autonomous campaigns. The harder questions concern agent access, defensive context and accountable control.

A contractor was removed after the FBI linked the FBIJobs.gov breach to an unapplied security patch, while key claims about the platform and scope remain source-attributed.

South Korea has expanded its response to financial-sector attacks, but reported ARTEX artefacts do not establish that an AI tool carried out an intrusion.

The incident used a private company’s authorised CPR connection, while investigators have yet to establish how much data was copied or whether it was misused.

Owners of working AT&T iPhone 18 Pro Max phones should check for iOS 27.0.1 and carrier settings 72.1. Official guidance says failed devices need replacement.

OpenAI’s dots combine persistent context, shared app permissions and action-taking tools. Its own documentation shows why disconnecting an app is not a complete data reset.

Investigators reportedly found temporary access to VL Prosperity’s propulsion technology, but the intruders’ capabilities, access route and identity remain unknown.

Fortinet reported active exploitation of CVE-2026-104286. Affected FortiMail operators need temporary controls and retrospective checks while verifying the availability of planned fixes.

RatHat’s credential-theft capabilities are documented, but they depend on sideloading, an Accessibility grant and abused debugging access.

Authorities seized KillSec’s leak site, servers and domains while arresting three suspects, including a minor alleged to have been a main operator.

Shinhan confirmed that sensitive information was extracted through a loan recruiter service, while reported evidence of AI involvement remains circumstantial.

ChatGPT for macOS build 26.924.20706 contains the documented fix, but the reported attack path began with code already running locally and no confirmed theft was identified.