AI is dramatically widening the cloud attack surface, with virtually every organisation seeing its AI apps and services under attack over the last year, according to Palo Alto’s latest report.

The cyber firm’s annual State of Cloud Security Report found that expanded cloud infrastructures, needed to host the influx of AI workloads, have become a critical target, leading to 99% of organisations experiencing an assault against their AI ecosystem in the past year.

At the same time, the rise of vibe coding (which the survey found was used by 99% of respondents) is generating insecure code faster than security teams can review it. 

Of the 52% of teams that ship code weekly, only 18% said they were able to fix vulnerabilities at the same pace, which the report said is leaving unaddressed risks that compound across cloud environments.

The explosion of AI has also opened new risk frontiers, with attackers pivoting to exploit foundational layers of cloud architecture, targeting API infrastructure, and identifying new ways to move laterally within networks. 

Particularly worrying, 41% of the 2,800 executives and IT pros polled reported a dramatic rise in attacks on APIs, the steepest surge of any threat vector this year, driven by ungoverned interface sprawl as AI agents become heavily integrated in workflows.

The flood of AI apps is creating cloud security blind spots, too. Firms report managing an average of seventeen cloud security tools from five vendors, leading to fragmented data and context gaps that slow incident response, a problem that might soon be solved, as 97% of organisations plan to consolidate their cloud security footprint.

Nearly half of organisations (47%) reported their biggest breach risk comes from data leaks via AI assistants and plugins, while threats like model supply chain tampering, endpoint abuse, and token theft are close behind.

Prompt injection and output manipulation also rank high, with the common thread across all these incidents being the exploitation of weak API boundaries, evidence which Palo Alto said highlights the risks posed by poorly controlled interfaces in AI environments.

However, despite the threat posed by the surge of AI applications across enterprise networks, identity remains the weakest link. 

Among those questioned, more than half (53%) cited lax identity and access management as a key challenge, with weak access controls emerging as a leading vector for credential theft and data leaks.

Recommended reading

• Are Security Teams Losing Visibility in the Cloud?
• Only 8% of Organisations Encrypt The Majority of Their Cloud Data
• Comment | Navigating Today’s Cloud Security Challenges

“As organisations aggressively scale cloud investments to power AI initiatives, they are inadvertently opening the door to sophisticated new attack vectors,” said Elad Koren, VP of product management for Palo Alto’s Cortex platform.

“Our research confirms that traditional approaches to cloud security are inadequate, leaving security teams to fight machine-speed threats with fragmented tools and slow, manual fix cycles. 

“Teams need more than just dashboards highlighting risks they can never burn down; they must transform with an agentic-first platform that spans code to cloud to SOC to finally operate faster than the adversary.”