1. Phishing & Social Engineering
Still #1 by volume and by consequence. AI-tailored phishing has roughly doubled click rates over generic blasts. The defense isn't training alone — it's phishing-resistant MFA (FIDO2, passkeys) so a clicked link doesn't equal a compromised account.
2. Credential Stuffing & Token Theft
Billions of leaked credentials are tested daily against login endpoints. Worse: session-token theft via infostealer malware bypasses MFA entirely. Controls that work: device-bound credentials, session-binding to client certificates, and short-lived tokens with refresh detection.
3. Internet-Exposed Vulnerability Exploitation
Public-facing servers with unpatched CVEs are still a top entry point. The window between disclosure and mass exploitation is usually under 72 hours. Asset inventory + an SLA on internet-exposed patches (24h critical) is non-negotiable.
4. Software Supply Chain
Compromised npm/PyPI packages, malicious GitHub Actions, and trojanized AI models. SBOMs and signed-artifact verification (Sigstore, in-toto) are the structural defenses; lockfile hygiene and pinning to immutable versions are the tactical ones.
5. Third-Party / Vendor Compromise
If your SaaS vendor is breached, your data may be too. Access tokens issued to vendors are a frequent pivot point. Vendor offboarding hygiene and scoped, expiring API tokens reduce blast radius.
6. Cloud Misconfiguration
Public S3 buckets, overly permissive IAM, exposed admin consoles. Most cloud breaches are configuration errors, not novel exploits. CSPM tooling and IaC scanning catch these before they reach prod.
7. Insider Threat
Malicious or negligent insiders. The negligent kind (clicked phishing, exfiltrated data via personal cloud) is far more common than the malicious kind. Controls: DLP, endpoint monitoring, segregation of duties.
8. Physical & Hardware Attacks
Lost or stolen laptops, USB drops, evil-maid attacks against unattended hardware. Full-disk encryption and Secure Boot are the table stakes.
9. IoT & Operational Technology
Cameras, printers, building-automation controllers, factory equipment. They run forgotten firmware, sit on the same VLAN as everything else, and rarely have logs you can centralize. Network segmentation is the single highest-impact control here.
10. Prompt Injection & LLM Abuse
Newer to the list but climbing fast. Indirect prompt injection (poisoned RAG content, malicious tool responses) lets attackers manipulate AI agents into leaking data or taking unauthorized actions. Tool gating, output schema enforcement, and treating retrieved content as untrusted are the structural defenses.