AI Security

AI Security
FBI and Amazon Warn ARTEX AI Tool Hunts Stored AI API Keys
Amazon and the FBI say attackers use the ARTEX AI pentest framework to find exposed AI apps and steal the provider API keys stored inside, ...
AI Security
ArcticRAT Android Malware Now Uses xAI’s Grok on Stolen Bank SMS
Zimperium found ArcticRAT Android malware calls xAI's Grok to spot banking apps and mine stolen SMS for OTPs and balances to size up each victim's ...
AI Security
AntiFox AI Pentest Platform Abused to Steal AWS Production Secrets
Arctic Wolf Labs links the AntiFox agentic pentest platform to attacks on IIS web apps and AWS environments that exposed production secrets and deleted S3 ...
AI Security
Over 35,000 Grok-Built Apps Found Open Online, Some Leaking Data
Claroc found 37,571 Grok-built apps online, 35,578 open without sign-in, some leaking staff and customer data from X, SpaceXSI and 30 other organizations.
AI Security
Check Point: ChatGPT HTML Preview Flaw Bypassed Lockdown Mode
Check Point Research says a patched ChatGPT HTML preview flaw escaped the sandbox and bypassed Lockdown Mode, enabling silent chat theft via custom GPTs.
AI Security
Researcher: Claude Dispatch Flaw Lets Browser Extension Run Code
Forever Security's Gal Weizman says a Claude Dispatch flaw lets a browser extension send commands to Claude Desktop and run code; Anthropic has not fixed ...
AI Security
OpenAI AI Pipeline Built 948 Linux Kernel Root Exploits
OpenAI says an agentic pipeline reproduced 1,724 Linux kernel flaws and built 948 working root exploits at about $1,279 each, in a run lasting 30 ...
AI Security
AI Tool ARTEX Used in Breaches at WorkOS and Chainstack
WorkOS and Chainstack disclosed intrusions tied to the ARTEX AI framework, exposing sales contacts, customer accounts and password hashes at the two firms.
AI Security
24,664 Lovable-Built Apps Leak Passports and Chat Messages
Crissy Field found 24,664 Lovable-built apps on Supabase returning database records with no authentication, exposing passports, payments and chat messages.
Hugging Face Datasets Flaw Lets Malicious Pickle Files Run Code
AI Security
Hugging Face Datasets Flaw Lets Malicious Pickle Files Run Code
The deprecated pandas builder can execute code from an untrusted pickle file. A proposed fix is still awaiting review