Serious vulnerabilities in cloud workloads have surged fourfold in 2026, according to Datadog's State of Cloud Security 2026 report, published on 10 November. The median number of high or critical flaws in a cloud workload rose from four to 14 in June and reached 16 by November.

Datadog, a cloud observability and security company, based the findings on analysis of the cloud security posture of thousands of customer organisations around the world. The figures come from vulnerabilities in third-party libraries used by applications monitored through its code security software.

AI is speeding up flaw discovery

Datadog links the rise to frontier AI. The increase coincided with the release of Anthropic's Mythos model in May, and Datadog describes the trend as a surge in reported vulnerabilities driven by security-capable AI models. Google Cloud estimates that 44.5% of cloud incidents in the first half of 2026 began with an attacker exploiting a software vulnerability.

Datadog says the growing number of flaws in the CVE database is not only a theoretical risk. "It affects an increasing number of real-world applications running in cloud environments, reinforcing the need for cloud security foundations that limit the blast radius of compromised applications," the report says.

Long-lived keys add to the risk

The report also highlights long-lived credentials, particularly for cloud-hosted AI services and software pipelines. More than one in five organisations Datadog analysed now use cloud-hosted AI inference services such as Amazon Bedrock, Azure Foundry Tools and Google Cloud's Gemini Enterprise Agent Platform.

Nearly half of organisations using Gemini rely on long-lived Gemini API keys, which stay valid until revoked. Almost 30% of AWS IAM users and Google Cloud service account users have access keys more than three years old. Datadog warns that such credentials are consistently among the most common causes of cloud breaches, and that automated tools scan the web for exposed keys at scale.

"That same tooling is available to autonomous attackers, who enumerate for long-lived credentials first because they're the highest-value, longest-lived prize in the environment," the report says. Datadog recommends using short-lived federated credentials, such as an IAM role or managed identity, instead of static API keys for AI services.

Dependencies adopted at machine speed

Cloud workloads carry an average of 149 software dependencies, and the average organisation has 568, Datadog found. Some 60% of organisations run dependency versions released less than 24 hours earlier, and those fresh versions appear in more than a fifth of cloud applications.

That speed matters because supply-chain attacks are increasingly run autonomously, with hijacked dependencies a popular route to compromise. Datadog recommends dependency cooldowns, so new versions are held back briefly while the community checks them, noting that hijacked libraries are usually identified and contained within a few hours.

Sources: Infosecurity Magazine and Datadog.