Skip to content
Back to Security Journey Blog

Five AI Security Risks Employees Need to Know and How to Prevent Them

Employees reviewing computer code, illustrating AI security risks and how to prevent them.
AI Enablement & Training AI Advantage

Published on

An untrained employee can accidentally create AI security risks when they're just trying to find a way to finish their work faster. With the right AI security awareness training, they can find ways to speed up their workflows without putting the company at risk.

The main AI security risks for employees to be aware of are shadow AI, sensitive data in prompts, unverified AI output, AI-generated phishing, and AI-generated code or content shipped without being reviewed first. You can prevent each of these risks with training aimed at the habit that causes it.

AI Security Risk 1: Shadow AI

Shadow AI is any AI tool employees use for work that their company hasn't approved or can't see. For example, an employee could paste a client email into a personal chatbot account to draft a reply. That AI provider would have no agreement with your company covering the email and your security team would never know it happened.

Your company can pull that work back into view with approved AI accounts, which makes shadow AI one of the easiest workplace AI risks to fix. The share of AI users on company-managed accounts rose from 25% to 62% between late 2024 and late 2025 at organizations Netskope monitors.1 You can follow that trend by:

  • Giving every team an approved tool for the work it already does with AI.
  • Posting the approved list somewhere employees can easily access.
  • Teaching employees why a personal account makes the same prompt riskier.

AI Security Risk 2: Sharing Sensitive or Confidential Data in Prompts

Your company loses control of any data an employee types or uploads into a public AI tool. University of Melbourne and KPMG researchers report that 48% of employees who use AI said they had uploaded sensitive company or customer information to public AI tools.2 About half of your AI users may have already done the same if your workforce resembles the people in that survey.

Teach each employee what counts as sensitive in their own work to lower these AI data security risks. Employees can replace names and account numbers with placeholders before they paste to protect sensitive information. Data loss prevention tools can also block sensitive data bound for AI websites.

AI Security Risk 3: Trusting AI Output Without Verifying It

AI tools sometimes invent citations, quotes, and figures that look completely real. These tools were designed to predict what a good answer looks like and don't always check their answers against real sources. They might sometimes hallucinate a paper and reference it to justify their own answer.

A 237-page review that Deloitte wrote for a government department cited academic papers that don't exist and included an invented quote from a court judgment.3 The department says Deloitte had to refund more than A97,000oftheroughlyA440,000 contract due to this mistake. Even a firm with a formal review process can send those errors to a client, so everyone needs a standard verification process.

Every employee who uses AI should open each source a tool cites and check every quote and number against the original as part of responsible AI use. They should treat any reference they can't find as fabricated until someone proves otherwise.

AI Security Risk 4: AI-Generated Phishing Attacks

Attackers now use AI to write convincing emails and copy real people's voices. A message can look and sound right and still come from an attacker.

Personalized phishing emails written entirely by AI got a 54% click rate in a peer-reviewed study of about 100 university participants.4 The AI emails matched emails written by human experts and drew more than four times the 12% click rate of generic spam. Any attacker with an AI tool could now send messages as convincing as an expert's.

The FBI reports that since April 2025, attackers have impersonated senior US officials with AI-generated voice and text messages.5 Your employees can protect themselves with these habits the FBI recommends:

  • Look up the person's number yourself and call them back.
  • Confirm any new number or messaging app through a channel you've already verified.
  • Keep one-time login codes to yourself no matter who asks.
  • Report every attempt to your security team.

AI Security Risk 5: Using AI-Generated Code or Content Without a Security Review

Review any AI output that goes to a customer or into production as carefully as work a person wrote. Ahrefs found that 7% of marketing and SEO professionals who use AI said more than three-quarters of what they publish is unedited AI output, up from 4% the year before.6 Those marketers are essentially publishing every mistake AI made.

That's why you need to fact-check AI-drafted content before it's published. Developers should put all AI-written code through the same review and security checks as any other code. Someone from another team should also review any script or automation a nontechnical employee creates before it has any access to company data.

How Can Organizations Train Employees to Avoid These Risks at Scale?

Train each habit within the employee's own role and measure if their behavior changed. Half of the Fortune 500 HR leaders Gallup surveyed in 2026 doubted their managers could coach employees on AI.7 A structured training program can teach what those managers can't.

People develop a sense of AI security awareness when they practice these habits on their own work. Security Journey's AI Advantage offers role-based AI training for employees, including a track for people who create apps with AI.

Security Journey assesses each person's skills on day one and tracks how far they've come. Employees who practice these habits every week can prevent AI security risks in the tools your company adopts in the future.

References

  1. Netskope Threat Labs. “Cloud and Threat Report: 2026.” January 2026. https://www.netskope.com/resources/cloud-and-threat-reports/cloud-and-threat-report-2026
  2. University of Melbourne and KPMG. “Trust, Attitudes and Use of Artificial Intelligence: A Global Study 2025.” April 2025. https://doi.org/10.26188/28822919
  3. Fortune. “Deloitte to partially refund Australian government for report with apparent AI-generated errors.” October 2025. https://fortune.com/2025/10/07/deloitte-ai-australia-government-report-hallucinations-technology-290000-refund
  4. Heiding, Lermen, Koo, Verdun, Schneier, and Vishwanath. “Evaluating Large Language Models’ Ability to Automate Spear Phishing.” Expert Systems with Applications. 2026. https://www.schneier.com/academic/archives/2026/02/evaluating-large-language-models-ability-to-automate-spear-phishing.html
  5. FBI Internet Crime Complaint Center. “Senior US Officials Impersonated in Malicious Messaging Campaign.” May 2025. https://www.ic3.gov/PSA/2025/PSA250515
  6. Ahrefs. “69% of Marketers Are Publishing More AI Content Than Last Year.” September 2026. https://ahrefs.com/blog/marketers-ai-content-creation
  7. Gallup. “AI’s Effect on Workplace Culture.” August 2026. https://www.gallup.com/workplace/712976/ai-effect-workplace-culture.aspx

Read more from:

and
SJ AI Access