Black Hat · November 13, 2026

ChatMate: test the full boundary from document input to interpreter host

video Featured · 75/100 practical score Prompt InjectionAgent SecurityAI Red Teaming
ChatMate: test the full boundary from document input to interpreter host video thumbnail
Why it matters

Ori Lahav’s ChatMate research traces a malicious document through Copilot’s code interpreter, sandbox privilege escalation and an Azure runtime host escape to an interactive prompt channel. The companion write-up identifies CVE-2026-32193 and explains how host egress bypassed pod-level network isolation. It reports fixes before the demo, whose attacker connection was simulated. This is a study of a disclosed chain, not evidence of a currently unpatched Copilot exploit.

My takeaway: Map document ingestion, interpreter permissions, shared mounts, host services and outbound traffic as one attack path. In an authorized test environment, verify that untrusted code cannot modify privileged startup files, reach host-management daemons or promote tool output into instructions. Check the relevant vendor remediation and capture both pod and host telemetry; network isolation at one layer does not establish isolation of the whole chain.
Keep exploring

More curated notes connected through Prompt Injection and Agent Security.

OpenAI News · framework

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 GenAI Security Project · guide

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.

OpenAI News · framework

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.