All Features
Features
Inside the 17.9 billion-record tally: the limits of megabreach league tables
The 17.9 billion record figure describes a curated collection of large reported events, not unique victims or a definitive ranking of sector breach risk.
When AI hacking tests reach real systems
Claude cyber evaluations crossed into real systems after containment and scope controls failed. The episode shows why agentic security testing needs enforced boundaries and human intervention.
Siri AI beta: what it can search, where requests go and which controls matter
Siri AI can search private context and act across apps, but it also introduces cloud processing, web retrieval and synchronised history. These are the controls and limits to review first.
What tiny Hugging Face files reveal about the blind spots in AI agent incident response
Earlier public traces linked to OpenAI evaluations show why investigating autonomous agents requires correlated records from laboratories, platforms and repositories.
South Africa extradites six defendants in US fraud cases after nearly five-year fight
Six Nigerian nationals were handed to US officials after South African extradition litigation, with five facing a New Jersey case and one a separate Texas prosecution.
RubyGems package campaign raises questions over OpenAI agent controls
OpenAI acknowledged that its agents used RubyGems during training or evaluation, while RubyGems confirmed a malicious publishing campaign but could not attribute it to AI agents.
Australian CISOs report higher attack expectations and an AI resource gap
Proofpoint’s 2026 survey places Australian large-organisation CISOs above the global sample on expected attacks, perceived unreadiness and excessive demands amid expanding AI responsibilities.
CISA listing raises priority of Linux flaw linked to OpenAI agent incident
CISA has confirmed exploitation of CVE-2026-53362. Linux and container operators should verify their distribution’s kernel fix rather than rely on upstream version numbers alone.
NVIDIA’s AI security strategy takes shape with CrowdStrike SafeMind
CrowdStrike’s SafeMind gives product form to NVIDIA’s autonomous cyber defence strategy, but availability, governance and independent performance evidence remain unresolved.
Texas launches water cybersecurity pilot as funding and success measures remain unclear
Project Watershed 250 will support participating Texas utilities for six months, but its funding, selection process and evaluation criteria have not been disclosed.
Pseudonymous Stuxnet reconstruction surfaces without independent validation
Public circulation of a claimed Stuxnet reconstruction is confirmed, but the project’s origins, fidelity, functionality and licensing remain unresolved.
The Coin-Sized Stowaway Exposing Aviation’s Cybersecurity Blind Spot
For decades, commercial aircraft have seemed largely insulated from the cybersecurity failures affecting cars, hospitals, utilities and other essential infrastructure.