Prompt Injection
CRITICALCrafted inputs override LLM system instructions to bypass safety controls or exfiltrate confidential data.
GenAI Security Intelligence
AI & LLM threat landscape · curated by Vikas Pandita, GenAI Security Specialist
8
AI Advisories
6
AI Exploits
35
AI News Items
AI Threat Categories
Deepfakes & Synthetic Media
HIGHAI-generated fake audio/video used for fraud, social engineering, and large-scale disinformation campaigns.
AI-Powered Phishing
HIGHLLMs generating highly personalised, grammatically perfect phishing at scale in any language.
Training Data Poisoning
HIGHCorrupting datasets to embed backdoors or malicious biases into production AI models before deployment.
Data Exfiltration via LLM
CRITICALExtracting sensitive PII or IP through crafted prompts against RAG systems and vector databases.
Rogue AI Agents
HIGHAgentic AI with excessive permissions taking unauthorised real-world actions without human oversight.
Model Theft
HIGHStealing proprietary AI models via systematic API extraction or direct model-weight exfiltration.
AI Fraud & Scams
HIGHVoice cloning and AI personas used for financial fraud, BEC, and executive impersonation at scale.
AI Risk for GCC / Middle East organisations
AI-Enhanced BEC & Wire Fraud
GCC banking sector facing AI-generated executive voice cloning targeting fraudulent wire transfer approvals.
Deepfake Government Communications
AI-forged official communications targeting GCC public sector — verify all digital correspondence and requests.
Arabic-Language AI Phishing
LLMs now produce fluent Arabic phishing — language quality is no longer a reliable detection indicator.
Smart City & ICS AI Attacks
AI-optimised attacks on Vision 2030 and UAE smart city infrastructure (SCADA + AI convergence risk).