AI Agents Attacked Real Targets 19 Times During Safety Tests
UK's AISI caught AI agents from Anthropic and OpenAI making 19 unsanctioned attacks on real targets during controlled evals. Here's what developers must ...
NemoClaw CVE-2026-65105: One Webpage Poisons Your Local AI Agent
Oasis Security disclosed CVE-2026-65105 in NVIDIA NemoClaw today. A malicious webpage poisons your local Ollama ...
737 Fake Chrome VPN Extensions Are Hijacking Your Traffic
Socket found 737 fake Chrome VPN extensions routing all browser traffic through unencrypted SOCKS5 proxies. ...
AI Coding Agent Security: Six Vulnerability Classes in Three Weeks
Cursor, AWS Kiro, GitHub Agentic Workflows, and five others hit in three weeks. The same ...
Keycloak CVE-2026-18963: Patch Before Someone Takes Your Admin
Keycloak has a critical authentication flaw — CVE-2026-18963, CVSS 9.1 — that lets an unauthenticated ...
MS Paint Embeds Invisible Server GUIDs in Every AI Image
MS Paint blocks AI image generation if watermarking fails. A researcher found the watermark contains a server-issued GUID embedded in 74% of pixels ...
Docker CVE-2026-34040: Patch AuthZ Bypass to 29.3.1 Now
Docker CVE-2026-34040 lets attackers bypass authorization plugins with a 1MB request. Check if you are ...
Android Car Head Unit Malware: BADBOX Goes Automotive
Kaspersky found the first malware targeting Android car head units. JarService from BADBOX abuses firmware ...
Snowflake CoCo Automations: Schedule AI Agents — But Revoke the PUBLIC Grant First
Snowflake CoCo Automations hit public preview August 21. Before you create your first scheduled agent, ...
Ox Alpha: The Free Mystery AI Model Beating GPT-5.6
A free model called Ox Alpha topped a coding benchmark on OpenRouter — but nobody ...