Barracuda's first-person incident account explains how a cloud access key left active after a vendor project became the entry point to a roughly 44-hour intrusion on November 1–2. The company says the key was apparently obtained after the AI vendor's development environment was compromised. From there, the attacker followed stored secrets across cloud accounts, analytics infrastructure, source-code access and Kubernetes. Short-lived repository branches triggered build workflows that exposed encoded secrets in job logs, and a container with a read-only host-filesystem mount still enabled credential harvesting. Barracuda attributes the observed tooling to ARTEX indicators but does not identify the operator. It found no direct evidence that customer integration credentials were accessed, yet insufficient logging prevented ruling that out; the response therefore treated reachable credentials as potentially compromised and contacted affected customers. This distinction matters: an observed access path and incomplete telemetry justify containment without proving every possible downstream theft occurred.
Barracuda intrusion: retire vendor keys and trace chained secret access
Related research
More curated notes connected through Agent Security and AI Red Teaming.
Pacing model development in an era of cyber-critical capabilities
OpenAI says preliminary evidence that Astra may meet its Critical cybersecurity threshold led it to pause frontier reinforcement-learning work for two weeks and keep its largest planned run on hold. New safeguards include stronger workload and network isolation, continuous boundary testing, token-level monitoring that escalates suspicious tool activity, and broader alignment checks for deception, reward hacking, and unauthorized access.
OWASP Top 10 for Agentic Applications for 2026
OWASP's community guide organizes agentic-system risk into ten categories, including goal hijacking, tool misuse, identity and privilege abuse, memory poisoning, insecure inter-agent communication, cascading failures, and rogue-agent behavior. It provides a shared taxonomy and mitigation starting point rather than a certification checklist or evidence that a deployed system is secure.
A blueprint for democratic governance of frontier AI
OpenAI proposes a three-part U.S. frontier-AI governance model: harmonize emerging state safety laws into a federal baseline, strengthen CAISI as an evaluation and standards institution, and coordinate a broader resilience program. Proposed controls include severe-risk evaluations, transparency reports, independent audits, safety-incident reporting, model-weight security, whistleblower protection, and periodic technical assessments.