Cybersecurity Trends in 2026: Shadow AI, Quantum & Deepfakes

High-level overview

• This is an end-of-year IBM Technology channel forecast and review of cybersecurity trends for 2026 (and beyond). It revisits prior predictions from 2023–2025 and uses data from IBM reports and other industry sources.

• Central theme:

  AI is both a force-multiplier for defenders and a powerful enabler for attackers.

• Other major topics covered: autonomous agents, deepfakes, quantum threats to cryptography, and practical authentication improvements (passkeys).

Key technological concepts, findings, and impacts

Shadow AI

• Unapproved or unsupervised AI instances in cloud environments increase breach cost and risk.
• IBM Cost of a Data Breach finding: organizations with shadow AI paid about $670,000 more per breach.
• About 60% of organizations lack AI governance/security policies to control shadow AI.

Deepfakes

• Rapid growth in cataloged deepfakes: ~500,000 in 2023 → ~8 million in 2025 (≈1,500% increase).
• Detection is becoming a losing battle as deepfakes improve. Recommended approach: train users to evaluate requests and actions rather than rely solely on artifact detection.

AI-generated attacks and malware

• AI and agents can generate exploits and polymorphic malware (self-changing to evade detection), lowering attacker skill requirements and making defense harder.
• Examples include automated ransomware chains, polymorphic malware creation, and automated kill chains that perform reconnaissance, exploit development, data theft, and ransom collection.

Agents (autonomous goal-driven AIs)

• Two principal threat vectors:
  • Attacks on agents: hijack or compromise agents to amplify damage.
  • Attacks by agents: malicious automation that executes attacks autonomously.
• Specific risks:
  • Rapid, amplified damage when agents are compromised.
  • Zero-click / indirect prompt injections (agent reads email and executes embedded instructions without user interaction).
  • Proliferation of non-human identities that require lifecycle and privilege management.
• Potential attacker use: hyper-personalized phishing, fully automated malware/ransomware campaigns, end-to-end automated kill chains, and enhanced social engineering using data and deepfakes.

Attack surface & LLM-specific vulnerabilities

• AI increases overall attack surface.
• OWASP’s Top Ten for Large Language Models ranks prompt injection as the top vulnerability (2023 and again in 2025).
• Defenses are needed that detect and mitigate prompt injection and other LLM-specific threats.

AI for defense

• AI is being used to detect and block AI-specific attacks (for example, an IBM-built product to detect prompt injections).
• Expect increasing adoption of adaptive, real-time AI-driven security tooling.

Quantum computing and post-quantum cryptography

• Quantum-capable systems will eventually break current public-key cryptography (the so-called “Q-Day” risk).
• Interest in quantum-safe / post-quantum algorithms is rising, but real-world deployments remain limited.
• Recommendation: begin implementing quantum-safe cryptography now.

Authentication & passkeys

• Passkeys (FIDO Alliance) are a practical, phishing-resistant alternative to passwords.
• FIDO report: about 93% of accounts with major providers (Amazon, Google, Microsoft, PayPal, etc.) are eligible for passkeys; roughly one-third of users have enabled them.
• IBM uses passkeys enterprise-wide; the presenter noted storing 17 passkeys in their password manager.
• Recommendation: adopt passkeys to reduce credential-based and phishing-related breaches.

Broader societal and industry impacts

• Education: stop trying to ban AI. Instead, teach students how to work with AI and AI-enabled workflows because workplaces will expect AI use.
• Arts, music, marketing, coding: AI will significantly change these fields (examples include AI-generated bands/music, AI-generated marketing copy, and AI-assisted code generation with fewer pure coding roles over time).

Data sources, tools, and studies referenced

• IBM Cost of a Data Breach Report (breach cost statistics and shadow AI findings).
• OWASP Top Ten for Large Language Models (prompt injection ranking).
• FIDO Alliance report on passkey eligibility and adoption.
• IBM product that detects prompt injections (example of AI defending against AI).

Practical recommendations

• Implement AI governance and security policies to control shadow AI.
• Deploy adaptive AI defenses to detect prompt injection and other AI-native attacks.
• Manage agent identities and privileges carefully (least privilege and lifecycle management).
• Train users to scrutinize requests coming from deepfakes or automated messages (assume deepfakes will keep improving).
• Begin planning and adoption of post-quantum (quantum-safe) cryptography now.
• Adopt passkeys to reduce credential and phishing risk.

Related content mentioned

• Annual IBM cybersecurity prediction videos (2023–2025).
• Prior video on zero-click attacks (indirect prompt injection through email agents).
• Upcoming video on AI’s impact on education (“AI in the future of education”).
• IBM Cost of a Data Breach Report (annual analytical source).

Main speakers and cited organizations

• IBM Technology channel presenter (IBM cybersecurity expert; also an adjunct professor at NC State University).
• Cited organizations and sources: IBM (Cost of a Data Breach Report; IBM product for prompt-injection detection), OWASP, FIDO Alliance, and industry adopters (Amazon, Google, Microsoft, PayPal, Target, TikTok).