#AI security

All articles tagged with "AI security"

NHS England wants existing and future software removed from public view by May 11.
Critics say hiding open-source code will not improve security against AI-assisted vulnerability discovery.
An open letter opposing the move has attracted hundreds of signatures.

DT Editorial Team·May 5, 2026·via newscientist.com

Anthropic is investigating reported unauthorized access to Claude Mythos Preview.
The company says it has found no evidence that its own systems were impacted.
The alleged access reportedly involved a third-party vendor environment.

DT Editorial Team·Apr 22, 2026·via techcrunch.com

Mozilla says Anthropic’s Mythos Preview helped find 271 vulnerabilities in Firefox 150 before release.
Firefox CTO Bobby Holley said the model was comparable to elite human security researchers.
The result intensifies debate over whether frontier AI will favor defenders or attackers in cybersecurity.

DT Editorial Team·Apr 22, 2026·via arstechnica.com

Anthropic is limiting access to Mythos through a select program called Project Glasswing.
Internal testing reportedly found the model could autonomously exploit vulnerabilities and access sensitive data.
Evaluators said Mythos identified serious Linux kernel flaws and constructed functional exploits.
The case points to a broader industry problem as frontier AI models gain stronger offensive cyber capabilities.

DT Editorial Team·Apr 17, 2026·via futurism.com

Researchers said they bypassed Apple Intelligence protections with prompt injection.
The issue was later corrected.
The episode shows on-device AI is not immune to model-manipulation attacks.

DT Editorial Team·Apr 10, 2026·via 9to5mac.com

DT Editorial Team·Apr 4, 2026·via artificialintelligence-news.com

DT Editorial Team·Mar 12, 2026·4 min read·via openai.com

DT Editorial Team·Feb 25, 2026·4 min read·via artificialintelligence-news.com