Generative AI enables attackers to operate at machine speed, allowing autonomous workflows that can exploit misconfigured Fortinet appliances, map internal networks, dump credentials from Active Directory, and extract data from thousands of organizations in under an hour. Picus proposes an agentic BAS architecture—four cooperating agents (Researcher, Red Teamer, Simulator, Coordinator)—that compresses threat intelligence to remediation into minutes and automates safe, high-coverage simulations and vendor-specific remediation. #Fortinet #Picus
Keypoints
- Generative AI enables attackers to scale autonomous attacks (initial access, internal mapping, credential dumping, data extraction) across thousands of organizations in minutes.
- Attackers embed LLMs into their kill chain via custom MCP servers for automated backdoor creation, vulnerability assessment, and prioritized execution of offensive tools.
- Defenders operate at “calendar speed” with manual handoffs and long incident response times, creating a decisive speed asymmetry versus machine-speed adversaries.
- Picus proposes an agentic BAS architecture with four agents (Researcher, Red Teamer, Simulator, Coordinator) that compresses the intelligence-to-remediation cycle to ~3 minutes while ensuring safe, reversible tests.
- The Agent Researcher ingests and structures CTI (e.g., CISA alerts) for machine consumption; the Agent Red Teamer creates safety-checked attacker recipes and blueprints from atomic tests.
- The Agent Simulator runs risk-free simulations across layers (endpoint, cloud, network, SIEM, EDR) producing proof-based detection/blocking evidence and high MITRE ATT&CK coverage.
- The Agent Coordinator translates findings into vendor-specific mitigations, opens tickets or triggers SOAR, and re-simulates to verify remediation effectiveness, enabling closed-loop validation.
MITRE Techniques
- [T1190] Exploit Public-Facing Application – Used to gain initial access by exploiting a vulnerable/misconfigured external appliance (‘leveraging a Fortinet misconfiguration’).
- [T1595] Active Scanning – AI-driven vulnerability assessment and internal mapping where LLMs automate discovery and assessment (‘internal network mapping via LLMs’).
- [T1046] Network Service Discovery – Internal infrastructure mapping to identify targets and services for lateral movement (‘internal network mapping via LLMs’).
- [T1003] Credential Dumping – Extraction of credentials from systems/Active Directory to escalate privileges (‘dumps credentials’).
- [T1078] Valid Accounts – Use of stolen or forged credentials to obtain higher-privilege access such as Domain Admin (‘gain Domain Admin access’).
- [T1105] Ingress Tool Transfer – Automated deployment/creation of backdoors and offensive tools onto compromised appliances (‘automated backdoor creation on compromised appliances’).
- [T1059] Command and Scripting Interpreter – Execution of offensive tools and scripted attack steps driven by AI-prioritized execution (‘AI-prioritized execution of offensive tools to gain Domain Admin access’).
- [T1041] Exfiltration Over C2 Channel – Data extraction and removal of sensitive data following compromise (‘ultimately data extraction’).
Indicators of Compromise
- [Configuration ] Fortinet appliance misconfiguration leveraged at scale – example: “Fortinet misconfiguration” used to gain initial access (no specific CVE or IP provided).
- [Threat intelligence alerts ] Sources ingested by the Agent Researcher – example: CISA alerts, news feeds (no specific alert IDs quoted).
- [Indicators referenced ] Generic IoCs and IoAs mentioned as defensive inputs – context: IoCs dumped into SIEMs and IoA rules pushed to EDR; no specific hashes, domains, or IPs provided.
- [Atomic tests / Test artifacts ] Enterprise atomic test library entries used for simulation – example: Picus Threat Library atomic tests (specific test names or file hashes not listed).