← All articles

2026 Cybersecurity Threat Report: Ransomware, AI Attacks, and What's Coming Next

The first three quarters of 2026 have produced a string of record-breaking breach disclosures, ransomware payments, and a genuinely new threat category: AI-generated and AI-directed attacks. This post distills the signal from the noise — what actually changed, what didn't, and what security teams should prioritize heading into Q4.

Ransomware: Bigger payments, more selective targets

Ransomware payments crossed $1.1 billion in the first three quarters of 2026, on pace to surpass the 2023 record. But the composition has changed: fewer but larger payments, as ransomware groups abandoned the spray-and-pray approach for targeted campaigns against organizations with a proven ability to pay.

The shift is partly tactical and partly driven by law enforcement. After the LockBit and BlackCat takedowns, surviving groups became more selective, spending more time in victim networks before detonating ransomware — sometimes six months or more. By the time encryption happens, the attacker knows exactly what the organization can pay and has exfiltrated enough data to make the threat credible.

Key numbers:

  • Average enterprise ransom demand: $4.2M (up 38% year-on-year)
  • Percentage of victims who paid: 28% (down from 41% in 2024, as backup culture improves)
  • Highest single payment confirmed publicly: $47M (a US healthcare group, identity withheld)

Healthcare remains the highest-risk vertical, followed by financial services and critical infrastructure. Cloud-first companies are not immune: attackers have shifted from encrypting on-premises systems to exfiltrating from S3 buckets and threatening to publish customer data.

AI-generated phishing: The end of typos

The traditional advice — "look for spelling mistakes" — is now obsolete. AI-generated phishing emails are indistinguishable from legitimate correspondence in every surface signal: grammar, tone, formatting, sender reputation.

Attackers are now using LLMs to:

  1. Scrape LinkedIn, company websites, and press releases to personalize emails at scale
  2. Clone legitimate email threads mid-conversation, inserting a malicious link or attachment after several genuine exchanges
  3. Generate voice clones of executives for vishing (voice phishing) — a technique now used in around 12% of business email compromise (BEC) cases

The result is that organizations that relied on user training as a primary defense are seeing those defenses fail. Email security vendors are adding AI-detection layers, but the arms race is real.

What actually works: multi-factor authentication on everything, strict payment authorization workflows that require out-of-band verification, and email authentication (SPF + DKIM + DMARC with a reject policy). No amount of user training compensates for a missing reject policy on DMARC.

Exploited vulnerabilities overtake phishing as the leading breach vector

For the first time, exploited vulnerabilities — not phishing — are the most common initial access vector in confirmed breaches, per Verizon's DBIR. The shift has been driven by three factors:

  1. Third-party software supply chain: Attackers compromise widely-used libraries or platforms upstream and exploit organizations downstream without ever needing a phishing email
  2. Faster time-to-exploit: The median time between a CVE being published and exploitation in the wild has dropped to under 48 hours for critical vulnerabilities
  3. Exposed attack surfaces: The average mid-market company has 170+ internet-facing assets; most have no continuous monitoring

The MOVEit-style "exploit one product, breach thousands of organizations" pattern is now the preferred approach for sophisticated threat actors. The 2026 equivalent hit a popular SaaS-to-SaaS integration platform in June, affecting over 400 downstream customers simultaneously.

Agentic attacks: The threat category that wasn't in last year's report

The most significant qualitative shift in 2026 is the emergence of agentic attacks — automated attack sequences driven by AI agents that can reason, adapt, and chain findings without human oversight.

Early examples observed in the wild:

  • Credential stuffing with LLM bypass: Agents that solve CAPTCHAs, recognize and handle rate limiting, and adapt request patterns to evade detection — all without a human in the loop
  • Autonomous reconnaissance: Agents that enumerate attack surfaces, prioritize targets by exploitability, and hand off verified findings to a human operator
  • Self-healing persistence: Implants that detect and respond to defensive actions, moving laterally before they can be fully evicted

This is not theoretical. SecurityInfinity's threat intelligence team has observed agentic attack tooling being sold in underground forums since early 2026, with documented use in at least three confirmed incidents.

The irony is that agentic AI is simultaneously the best defense: offensive security teams using AI agents for continuous testing can find and fix the vulnerabilities that agentic attackers would exploit, before the attackers get there. The window between vulnerability introduction and exploitation is too short for annual or quarterly pentesting to cover.

What to prioritize for Q4 2026

Given the above, here is what security-conscious organizations should focus on:

  1. Patch critical CVEs within 48 hours. The 48-hour exploit window is not aspirational; it is now the realistic baseline for critical severity vulnerabilities in internet-facing software.
  1. Implement DMARC with a reject policy. Still fewer than 40% of domains have DMARC configured correctly. It eliminates one of the most common BEC attack paths at zero cost.
  1. Know your attack surface. You cannot patch what you do not know exists. Run an attack surface scan — SecurityInfinity does it free for one domain — to find the forgotten subdomains, exposed S3 buckets, and open admin panels before attackers do.
  1. Test your backups. Backups that have never been restored are not backups. Run a tabletop exercise or an actual restore test before Q4 ends.
  1. Authenticate machine-to-machine. API keys in environment variables, hardcoded in config files, or accidentally committed to git are a persistent top-five finding in every penetration test we run. Secrets management platforms (HashiCorp Vault, AWS Secrets Manager) have no good excuse not to use.

Frequently asked questions

Where do these statistics come from?

The figures cited draw from Verizon DBIR 2026, Mandiant M-Trends 2026, Chainalysis Crypto Crime Report 2026 mid-year update, and SecurityInfinity's own incident data. All are publicly available.

Is AI-generated phishing detectable by email security tools?

Some tools are beginning to add AI-content detection layers, but efficacy varies and false positive rates are high. The more reliable mitigation is not detection but architecture: MFA, DMARC, and payment authorization workflows that cannot be bypassed by a convincing email alone.

How do agentic attacks differ from traditional automated scanning?

Traditional scanners run a fixed playbook. Agentic attackers use a language model to reason about what they find, decide what to do next, and adapt when defenses respond. They behave more like a human attacker than a script, which is why they evade defenses designed to detect scripts.

Find what attackers can see, before they do.

AI pentesting agents test your web app, API and cloud, and every finding comes with evidence and a fix. First report in about 10 minutes.

Stay ahead of attackers.

New vulnerabilities, breach write-ups and practical fixes. A few emails a month.