Cloud Security Statistics 2026: How Cloud Environments Get Breached

Cloud breaches used to be a story about weak passwords and open storage buckets. The 2026 data says that story has changed: unpatched third-party software running in the cloud is now the leading entry point, credential theft is still rising, and the gap between a vulnerability being published and being exploited at scale has collapsed to days. This page collects the measured figures from Google Cloud’s Threat Horizons report, CrowdStrike’s Global Threat Report, the Thales Data Threat Report, IBM’s Cost of a Data Breach and Verizon’s DBIR, each linked to its source. We update the page as each is published.

How to cite: link to this page or to the primary source beside each figure. Survey figures carry their sample size.

1. How cloud environments are breached

• Exploitation of vulnerabilities in third-party software was the primary entry vector in 44.5% of cloud intrusions in the second half of 2025, up from 2.9% in the first half: the first time it overtook stolen credentials since the report began. (Google Cloud Threat Horizons Report, H1 2026)
• Weak or absent credentials fell from 47.1% of entry vectors to 27.2% over the same period. (Google Cloud Threat Horizons, H1 2026)
• The window between a vulnerability’s public disclosure and mass exploitation in the cloud collapsed by an order of magnitude, from weeks to days. React2Shell (CVE-2025-55182) was exploited within 48 hours of disclosure in December 2025, with cryptominers deployed at the same speed. (Google Cloud Threat Horizons, H1 2026)
• Cloud-conscious intrusions rose 37% in 2025, and cloud targeting by state-linked actors rose 266%. (CrowdStrike 2026 Global Threat Report)
• 67% of organizations see credential theft and misappropriated secrets increasing against their cloud infrastructure. (Thales 2026 Data Threat Report, 3,120 respondents)
• 48% of all breaches involved a third party, including SaaS providers, cloud platforms and OAuth integrations, a 60% increase in one year. (Verizon 2026 Data Breach Investigations Report)
• Exposed applications and systems were the entry point in 38% of ransomware attacks, ahead of user devices (30%), firewalls (21%) and VPNs (8%). (Sophos State of Ransomware 2026, 2,158 organizations)

What this means for testing: a cloud penetration test that only reviews IAM policies misses the vector that now leads. The scope has to include every internet-facing workload and the software running on it. See cloud penetration testing for how we scope AWS, Azure and GCP engagements.

2. Identity in the cloud

• More than 97% of identity attacks are password attacks, and identity-based attacks rose 32% in the first half of 2025. Phishing-resistant MFA blocks over 99% of them. (Microsoft Digital Defense Report 2025)
• 79% of ransomware attacks began with an identity-based approach; 97% of victims whose credentials were stolen had MFA enabled somewhere, but not on the account that was used. (Sophos 2026)
• Credential abuse appeared somewhere in 39% of breaches even though it was the initial vector in only 13%. (Verizon 2026 DBIR)
• The average eCrime breakout time, from a first foothold to lateral movement, fell to 29 minutes; the fastest was 27 seconds. In cloud environments that movement is an API call, not a network hop. (CrowdStrike 2026)

3. Data in the cloud

• Only about half of sensitive data stored in the cloud is encrypted: 47%. (Thales 2026)
• Only 34% of organizations say they know where all their data resides, and 39% can fully classify it. (Thales 2026)
• Human error was named the leading cause of breach by 28% of organizations. (Thales 2026)
• Attackers sought to steal data in 80% of the incidents Microsoft’s security teams investigated. (Microsoft Digital Defense Report 2025)
• Breaches through the supply chain took the longest to identify and contain, 258 days against 247 for all breaches. (IBM Cost of a Data Breach Report 2026, 602 organizations)

4. AI workloads, the new cloud attack surface

• More than 20% of organizations reported a breach targeting their AI models or applications. The leading causes were compromised APIs, applications or plug-ins (27%) and cloud misconfigurations affecting AI workloads (27%). (IBM 2026)
• 61% of organizations rank AI as their top data security risk, and 61% say their AI applications are being targeted, with sensitive data the leading target. (Thales 2026)
• Shadow AI incidents affected 43% of breached organizations, up from 20% a year earlier; employees using unapproved AI tools tripled from 15% to 45%. (IBM 2026; Verizon 2026 DBIR)
• AI-enabled breaches averaged $6 million, about $1 million above the overall average, and were one in four malicious breaches. (IBM 2026)
• AI-enabled adversary operations increased 89% year over year. (CrowdStrike 2026)

Our AI and LLM penetration testing service covers the model, the application around it, and the cloud plumbing underneath.

5. The cost of getting it wrong

• The global average breach cost $4.99 million in 2026 and $11.5 million in the United States. Financial services averaged $6.3 million. (IBM 2026)
• Mean time to identify and contain a breach was 247 days; breaches over 200 days cost $5.65 million against $4.32 million for shorter ones. (IBM 2026)
• Only 26% of known exploited vulnerabilities were fully remediated in 2025, down from 38%, and the median time to remediate rose to 43 days, which is the same window attackers now close in 48 hours. (Verizon 2026 DBIR)
• 42% of vulnerabilities exploited by adversaries in 2025 were exploited before public disclosure. (CrowdStrike 2026)

6. What the numbers say to do

1. Inventory every internet-facing workload and the software on it. The leading cloud entry vector is now a known vulnerability in third-party software. A cloud penetration test starts with that inventory, from outside, the way an attacker would.
2. Assume any credential can leak. 67% see credential theft rising. Test what a leaked developer key or an over-permissive role can reach; that assumed-breach scenario is the second half of our cloud engagement.
3. Patch on the attacker’s clock, not the audit’s. 48 hours to exploitation against a 43-day median to remediate is the whole problem in two numbers. Prioritize by exposure, not by CVSS alone.
4. Test the AI stack like any other application. 27% of AI breaches came through the APIs and plug-ins around the model, and another 27% through cloud misconfiguration. Both are ordinary findings in an API penetration test.

Our AWS penetration testing guide covers what AWS allows, how to scope it, and what gets tested.

Sources

• Google Cloud, Threat Horizons Report, H1 2026
• CrowdStrike, 2026 Global Threat Report
• Thales, 2026 Data Threat Report
• IBM, Cost of a Data Breach Report 2026
• Verizon, 2026 Data Breach Investigations Report
• Sophos, The State of Ransomware 2026
• Microsoft, Digital Defense Report 2025