IBM Turbonomic Operator Grants Cluster-Wide Access to Secrets
Prometurbo's service account could read secrets outside its own namespace. Reinstall the fixed agent and test the operator's effective permissions.
Read article →AI
AI systems, model security, agent abuse, and attacks that use generative tools.
Kotaemon Lets Signed-In Users Read and Delete Other Users' Chats
An ownership check is missing from four multi-user chat handlers through v0.12.0. Limit account access and review cross-user conversation activity while awaiting a verified fix.
Read article →
GTIG Counts 141 Exploited Flaws as CVE Disclosures Double
GTIG recorded 141 disclosed and exploited vulnerabilities through August 2026. Prioritize reachable edge systems and verify the access conditions on affected AI gateways.
Read article →
RubyDoc Builds Let Malicious Gems Run Code on Shared Servers
More than 100 malicious gems used RubyDoc documentation builds to execute supplied code. Maintainers should review package and account changes tied to legacy API keys.
Read article →
Cloud Intruders Automate Mass Credential Harvesting in Under Six Hours
Google observed attackers move from cloud compromise to agent-enabled mass credential harvesting in under six hours, shrinking the time defenders have to respond.
Read article →
Claude Ports WAGO PLC Exploit and Bricks Test Controller
Forescout ported a pre-auth PLC exploit with Claude, then bricked the device during an implant attempt. Restrict FTP and monitor crashes and outbound traffic.
Read article →
LiteLLM Flaw Lets Read-Only Accounts Steal AI Gateway Secrets
Zenity saw file-read probes matching CVE-2026-35029. LiteLLM advises upgrading to 1.83.0 or later; defenders should hunt configuration changes and rotate exposed secrets.
Read article →
AI Agents Help Ransom Attacker Seize Enterprise Root Credentials
Unit 42 traced an AI-assisted ransom intrusion across web, repository, secrets, CI/CD and cloud systems in less than 10 hours. These behaviors can reveal the loop.
Read article →
Chinese-Speaking Operator Uses AI Agents in Government and Education Intrusions
Five exposed SecFlow workspaces linked Claude, Qwen and DeepSeek to Asian government and education intrusions involving credential theft, webshells, and implants.
Read article →
Leaked AWS Administrator Key Funds Unauthorized AI Model Access
Fortinet traced one stolen AWS administrator key through a new IAM user, Marketplace agreements, and billable Bedrock model calls.
Read article →
Attackers Used METR’s Exposed API Key for Credits Worth $600,000
A fail-open agent dashboard exposed one public-model key for three weeks, while noisy usage signals and absent spend limits delayed recognition.
Read article →
Malicious Repositories Make Amazon Kiro Send Local Secrets Off-Device
A crafted workspace can steer Kiro 0.7.45 from reading a local secret to placing it in a Powers registry request. Amazon fixed the reported behavior in 0.8.140.
Read article →
Exposed Server Reveals Aurora Operator's Breach Playbook Across 17 Organizations
An exposed server revealed a repeatable Active Directory attack path, AI-assisted planning, Aurora lockers, and payment trails across multiple victims.
Read article →Showing 14–26 of 47