Cisco-State-of-AI-Security-2026.md

The State of AI Security 2026 report reviews how AI has become a high-risk attack surface: adversaries (including nation-state APTs and criminal groups) are weaponizing agentic AI, exploiting Model Context Protocol (MCP) and agent-to-agent (A2A) ecosystems, poisoning training data and model supply chains, and leveraging jailbreaks/prompt injection to exfiltrate data or automate attacks. The report documents concrete incidents and vulnerabilities (CVE disclosures, malicious npm packages, LAMEHUG malware using LLMs, Anthropic/Claude compromise), provides a unified AI security taxonomy and tooling (MCP/A2A scanners, Skill Scanner, pickle-fuzzer), and urges defense-in-depth, lifecycle controls, and supply-chain hardening to mitigate large-scale AI-enabled threats.