Companies are facing new AI-related cybersecurity risks as employees are using many AI tools, as well as rogue browser extensions and vulnerable autonomous agents that have exposed organisations to new types of cyberattacks, according to an Akamai report.
While AI use has grown across business functions, it revealed, much of the risk is concentrated among a small group of highly active “AI power users” who account for a large share of enterprise AI exposure.
Nearly half or 47.11 per cent of all enterprise AI conversations occur via personal identities, instead of corporate-managed accounts (52.89 per cent).
This “shadow AI” is activity that has not been approved, monitored or audited by IT, security, or compliance teams, according to Akamai, which provides content delivery and security services to businesses.
Its report also found that corporate AI adoption has shifted sharply since early 2025, moving from cautious experimentation to becoming a structural part of business operations. However, Akamai warned that this integration has outpaced traditional security controls.
“AI is no longer just a productivity booster; it is a collaborative colleague with direct access to the corporate crown jewels,” said Or Eshed, the vice-president for enterprise security product and engineering at Akamai.
He added that traditional data loss prevention (DLP) tools were built for an earlier era of file transfers and e-mails, while sensitive corporate information is now being broken up across prompts, personal accounts and autonomous AI agents.
“Security leaders must pivot from trying to block AI to continuously governing how it operates at the interaction level,” he advised.
New AI-native threats
The report highlighted three emerging AI-native attack methods discovered by researchers in 2026, which Akamai said can bypass traditional perimeter defences.
The first, described as vibe hacking, involves attackers covertly changing local markdown instruction files in a developer’s environment. These modifications can trick AI coding assistants into generating insecure code or carrying out unauthorised actions while appearing to follow the developer’s normal workflow.
The second, called CursorJacking, involves rogue browser extensions using broad permissions to silently collect API keys, proprietary codebases and conversation history from the browser environment. Akamai said this technique targets users of popular AI coding assistants, including Cursor.
The third, CometJacking, uses indirect prompt injection. Attackers embed malicious instructions on a public webpage, which can then manipulate a user’s local AI agent.
According to Akamai, a compromised agent may be able to exfiltrate local files, e-mails and session credentials without the user’s knowledge. The report said this threat targets agentic browsers such as Perplexity’s Comet AI.
Strategies to secure AI
For organisations to gain the benefits of AI without exposing critical data, the report suggests five mitigation strategies for chief information security officers (CISOs).
Businesses should first focus on high-risk AI power users, directing monitoring, telemetry and tailored coaching toward the small group of employees driving the majority of interactive AI prompts, according to Akamai.
It also called on companies to reduce shadow AI by enforcing single sign-on (SSO) federation across platforms and continuously discovering the long tail of niche AI software-as-a-service tools being used within the organisation.
Security teams, it said, should move beyond static DLP and inspect the AI interaction layer in real time, including prompts, copy-and-paste buffers and document uploads.
Organisations are also urged to treat browser and integrated development environment extensions as highly privileged software. According to Akamai, almost 75 per cent of AI extensions request high or critical permissions, while 16.3 per cent contain known vulnerabilities.
Finally, the company said businesses need to secure autonomous AI agents by setting strict least-privilege boundaries and introducing behavioural monitoring for agents that act on behalf of employees.
