Akamai: 5% of AI Users Create Outsized Risk
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Akamai published its State of the Internet: Enterprise AI Usage Risk Report 2026, finding the top 5% of enterprise AI power users interact with AI at 12 times the rate of the bottom 50% and average 18+ prompts per conversation versus roughly five for typical employees.
- Akamai's report found 47.11% of enterprise AI conversations flow through personal rather than corporate identities, with DeepSeek (99.8%), Microsoft Copilot Standard (63.92%), ChatGPT (61.36%), and Claude (61.09%) dominated by personal logins, while Gemini Enterprise (98.15%) and Microsoft Copilot M365 (90.55%) keep most interactions inside corporate identity systems.
- 14.4% of enterprise AI conversations use corporate email addresses tied to personal "freemium" subscriptions, meaning sensitive data injected into prompts may feed public model training.
- Browser and IDE AI extensions represent a major blind spot, used by 17.7% of midsize enterprise employees (vs. 9.53% at larger organizations), with nearly 75% requesting high or critical permissions and 16.31% carrying known CVE vulnerabilities compared to 10.80% of browser extensions overall.
- Akamai identified three emerging attack vectors — Vibe Hacking (modifying local instruction files like AI_CONFIG.md to manipulate coding assistants), CursorJacking (rogue extensions harvesting API keys and source code from local databases), and CometJacking (indirect prompt injection via malicious web pages to trick AI agents into exfiltrating user files).
- Or Eshed, Akamai's VP of Enterprise Security Product & Engineering, said small groups of AI power users are "casting outsized shadows across enterprise threat surfaces" and called AI "a virtual colleague with keycard access to the company vault."
Why it matters: CISOs focused on governing ChatGPT are missing the real exposure: 47.11% of enterprise AI conversations flow through personal identities, and 14.4% use corporate emails tied to freemium accounts where prompt data may feed public model training. With 16.31% of AI extensions carrying known CVEs, attackers now have direct paths into enterprise AI agents.
Ask SkimNews
