AI exploits double software vulnerabilities
⚠️ Software vulnerabilities reached roughly 85,000 in 2026, doubling the previous year's total as neural networks generate exploits.
Flaws actively used in cyberattacks also doubled in frequency.
#cybersecurity ⚡ #vulnerabilities ⚡ #infosec
Subscribe
• Sent via TeleFeed
⚠️ Software vulnerabilities reached roughly 85,000 in 2026, doubling the previous year's total as neural networks generate exploits.
Flaws actively used in cyberattacks also doubled in frequency.
#cybersecurity ⚡ #vulnerabilities ⚡ #infosec
Subscribe
• Sent via TeleFeed
An internal AI model autonomously chained zero-day exploits
An unreleased internal version of Anthropic's model autonomously discovered and chained zero-day vulnerabilities across Chrome, Safari, Android, Windows, and Linux on October 28. Claude successfully bypassed browser sandboxes, extracted photos from a locked Google Pixel 9, and executed root-level commands on servers without human intervention.
⚠️ During testing, the AI found 62 high-severity flaws, including 13 ways to escape the Chrome sandbox, allowing a webpage to launch system applications in seconds. Similar multi-step attack chains forced immediate patches from Google, Apple, and Microsoft before public disclosure.
The system required physical access to target hardware for the Android data extraction, yet bypassed screen locks and data transfer restrictions automatically.
#CyberSecurity ⚡ #ArtificialIntelligence ⚡ #Vulnerabilities
Subscribe
• Sent via TeleFeed
An unreleased internal version of Anthropic's model autonomously discovered and chained zero-day vulnerabilities across Chrome, Safari, Android, Windows, and Linux on October 28. Claude successfully bypassed browser sandboxes, extracted photos from a locked Google Pixel 9, and executed root-level commands on servers without human intervention.
⚠️ During testing, the AI found 62 high-severity flaws, including 13 ways to escape the Chrome sandbox, allowing a webpage to launch system applications in seconds. Similar multi-step attack chains forced immediate patches from Google, Apple, and Microsoft before public disclosure.
The system required physical access to target hardware for the Android data extraction, yet bypassed screen locks and data transfer restrictions automatically.
#CyberSecurity ⚡ #ArtificialIntelligence ⚡ #Vulnerabilities
Subscribe
• Sent via TeleFeed