Cloud vulnerabilities quadrupled in nine months, Datadog reports

The median number of high or critical vulnerabilities in the cloud services Datadog analysed rose from four in February 2026 to about 16 in November, a fourfold increase in nine months. Datadog published the figures on 11 November 2026 in its State of Cloud Security 2026 report, which draws on customer data from thousands of organisations worldwide using its infrastructure monitoring, logging and cloud security products.

The stakes extend beyond any single application. The report argues the central question in cloud security has shifted: it is no longer only whether an application can be compromised, but how far an attacker could move after gaining access.

How exposed are AI access keys?

More than one in five organisations analysed uses cloud AI services, according to the report. Among those using Google Gemini, nearly half rely on long-lived Gemini API keys. Among organisations using Amazon Bedrock, 17 per cent use long-lived Bedrock API keys.

Those credentials remain valid until revoked. A stolen key can continue to grant access long after the initial breach, and the exposure grows as businesses build AI tools into internal systems and customer-facing applications.

Andrew Krug, Head of Security Advocacy at Datadog, warned that AI access keys give attackers a practical shortcut. "Attackers don't need to use their own AI model if they can steal the key used to access yours," he said.

Datadog's Security Labs team recently identified tools built to test whether stolen Amazon Web Services credentials could reach Amazon Bedrock. Researchers observed similar probing activity in 12 organisations' cloud environments over a 30-day period.

What is driving the software supply chain risk?

Software package dependencies form a second front. Cloud workloads depend on 149 software packages on average, and more than half of those are indirect dependencies introduced through other packages, the report found.

Deployment speed compounds the exposure. Sixty per cent of organisations run cloud applications using a package version released less than 24 hours earlier. Across all cloud applications analysed, 21 per cent use a version that new.

Datadog cautions that recently released packages are not inherently unsafe. The finding instead shows how quickly a compromised component could spread into production systems. The report cites Shai-Hulud 2.0, malware that searched infected systems for cloud credentials, as an example of the pattern.

The report frames these findings as part of a wider shift in security risk, in which application flaws, software supply chain issues and machine credentials intersect. An attacker who gains access through vulnerable software or a poisoned package can then use inherited permissions and static keys to move across systems.

Are defenders making progress?

Some indicators point in the right direction. Among organisations using AWS, 55 per cent have adopted data perimeters, up from 40 per cent in 2025. Data perimeters narrow what an attacker can reach after an initial breach.

Emilio Escobar, Chief Information Security Officer at Datadog, said remediation alone no longer suffices. "As serious flaws multiply, the question is how far an attacker can get through one compromised application," he said. "Teams need to fix those flaws and limit the damage if an application is compromised. Temporary credentials and data perimeters can restrict an attacker's access, while a shared view of applications, identities and infrastructure helps security and engineering contain an attack before it spreads. In the AI era, security becomes an observability problem: you can't stop a system-wide attack without seeing the whole system."

The report recommends replacing static keys with temporary credentials where possible. It singles out AI deployments, where service-to-service access is often granted early in development and then left in place. Convenience in deployment, the report concludes, is still outweighing risk reduction in many environments.

Is this a global pattern?

Datadog's regional data indicates the problems are not confined to one market. Yadi Narayana, Field CTO APJ at Datadog, said the company sees similar issues in Australia and New Zealand, and that long-lived credentials persist for predictable reasons.

"The report broadly reflects what Datadog sees across Australia and New Zealand. Many organisations understand the risk, but long-lived credentials often remain hidden in older applications and integrations because teams fear disrupting critical services. Boards should treat software supply chain risk as a business resilience issue, not simply a technical problem. The priority is to identify every credential, understand what it can access, replace static keys where possible and monitor unusual activity across the full environment," Narayana said.

For security teams, the report's arithmetic sets the immediate agenda: four times as many serious vulnerabilities as in February, roughly 149 dependencies per workload, and static AI keys in place at a substantial share of Gemini and Bedrock users. The organisations that map every credential and constrain what each one can reach will determine whether a quadrupled vulnerability count translates into quadrupled breach impact.