CSA's 2026 Top Threats: Identity Beats Infrastructure, AI Shows Up Twice — Malay Tech JournalSkip to content
ENMS
EssayGround Truth

CSA's 2026 Top Threats: Identity Beats Infrastructure, AI Shows Up Twice

Cloud Security Alliance ranks inadequate IAM — driven by non-human identity sprawl — as the #1 cloud threat for 2026, with two AI-related risks debuting on the list. What actually changed, and what it means for cloud teams.

5min read

Every year, the Cloud Security Alliance polls its Top Threats Working Group and a few hundred industry practitioners, and ranks what’s actually worrying cloud security people — not what vendors want to sell them. The Top Threats to Cloud Computing Survey Report 2026, published 13 August 2026, just reordered the list in a way worth stopping for.

Inadequate Identity and Access Management is now #1. It was #2 in the 2024 report. The threat that held #1 in 2024 — Misconfiguration and Inadequate Change Control — dropped to #5. And for the first time, two AI-related threats made the list at all.

This isn’t a reshuffle of ten similar items. It’s CSA saying, in effect: the thing most likely to hurt you in 2026 isn’t a misconfigured S3 bucket. It’s who — or what — is allowed to log in.

CSA’s own framing is direct: “The rapid growth of non-human identities, machine-to-machine interactions, and autonomous decision-making has now surpassed what traditional management and oversight processes were designed to handle.”

Read that again without the corporate phrasing: most IAM programs were built to manage humans with passwords, and now most of the identities in a cloud environment aren’t human. Service accounts, CI/CD pipeline credentials, API keys, workload identities, and — increasingly — autonomous AI agents acting on a schedule or in response to events. None of them take a lunch break, none of them get suspicious about an unfamiliar login prompt, and most of them were provisioned once and never revisited.

That’s not a hypothetical. It’s the same root cause behind the identity-first attacks already covered on this site — Snowflake, MGM, and Scattered Spider didn’t need an exploit. They needed a valid credential. CSA’s 2026 data says the industry now agrees that’s the biggest exposure, not an edge case.

Two new entries appear in the 2026 ranking, both AI-related, both new:

  • AI-Enhanced Attacks (#2) — AI used to improve or automate attacks themselves: faster reconnaissance, more convincing phishing, adaptive malware.
  • AI System Compromise (#6) — AI systems treated as assets in their own right, which can be compromised, manipulated, or abused — think poisoned training data, prompt injection against agentic pipelines, or a hijacked inference endpoint.

Vic Hargrave, co-chair of CSA’s Top Threats Working Group, put it plainly: “AI is not an emerging cloud security concern. It is already changing both how attacks are carried out and what organizations have to protect.” Two threats on an 11-item list is a small number, but it’s the first time AI has appeared in this survey at all — and it entered swinging, taking the #2 slot outright.

It’s worth being precise here, because two different things happened to “traditional infrastructure” concerns, and they’re easy to conflate:

  1. Misconfiguration and Inadequate Change Control — 2024’s #1 threat — didn’t disappear. It fell to #5. Still a real, ranked concern, just no longer the top one.
  2. A separate category covering cloud infrastructure and cloud service provider risk, which existed on the 2024 list, scored low enough in 2026 that CSA dropped it from the list entirely.

Different mechanisms, same direction: the industry’s attention moved from “is the platform itself risky” toward “who has access to it, and is that access being watched.”

#Threat
1Inadequate Identity and Access Management
2AI-Enhanced Attacks
3Insecure Third-Party Resources
4Insecure Interfaces and APIs
5Misconfiguration and Inadequate Change Control
6AI System Compromise
7Advanced Persistent Threats
8Lacking Cloud Security Strategy and Governance
9Insecure Software Development
10Accidental Cloud Data Disclosure
11System Vulnerabilities

The methodology behind it: a two-stage process, starting with an in-person survey of the Top Threats Working Group, followed by a poll of 500+ industry practitioners rating 23 candidate issues on a 10-point scale. It’s a perception survey of practitioners, not a breach-frequency dataset — worth keeping in mind when reading the ranking as “what’s most dangerous” versus “what’s most top-of-mind.”

CSA’s report doesn’t mention Malaysia specifically — this part is analysis, not something the report says. But the underlying pressure applies directly here: Malaysian cloud estates (banks and GLICs under Bank Negara’s RMiT expectations, and the broader government cloud-first push) are accumulating non-human identities — CI/CD service accounts, API integrations, and now AI agents — faster than most teams are auditing them.

Two concrete things worth doing before this shows up as an incident instead of a survey ranking:

  • Inventory non-human identities the same way you’d inventory human accounts. Service accounts and API keys that were provisioned once, granted broad scope “to be safe,” and never rotated are exactly the gap CSA is describing.
  • Put AI agents and pipelines inside your existing IAM and logging model — don’t treat them as a special case. If an AI agent can call an API or trigger a deploy, it needs the same least-privilege and audit trail as a human with those same permissions.

The headline isn’t “AI is scary” or “identity is hard” — both were already true. The headline is that CSA’s own practitioner survey now ranks identity, specifically non-human identity, above the infrastructure and configuration issues that dominated the conversation for years. If your 2026 cloud security roadmap still leads with misconfiguration scanning and hasn’t got a line item for machine identity lifecycle management, this report is a good reason to reorder it.

Sources: