A Fifth of UK Workers Share Passwords and Credentials With AI Tools
Share
Nearly one in five UK employees have shared sensitive login credentials, including passwords and API keys, with artificial intelligence (AI) tools, according to new research into workplace security habits. The findings highlight a growing trend where staff extend excessive trust to unmanaged AI agents, often bypassing established corporate security protocols.
A study of 1,000 UK workers by credential security firm Dashlane found that 19% of employees have provided login details or API keys to an AI tool. Furthermore, nearly half of the respondents (48%) admitted to allowing AI systems to perform tasks such as accessing company emails, internal files, and customer databases without first seeking explicit permission from their IT departments.
The Rise of the AI Colleague Effect
Researchers have identified an emerging “AI Colleague Effect,” where employees treat AI agents as trusted coworkers rather than third-party software. This misplaced trust leads to risky behaviours; approximately 37% of workers said they would be willing to provide login credentials to a new AI tool after using it for just one week.
The data suggests that the use of “shadow AI”—the deployment of AI tools without official authorisation—is widespread. While 92% of IT executives surveyed claimed to have a complete overview of every AI agent operating within their organisation, 25% of employees admitted to using AI tools that their employers are unaware of. This visibility gap creates significant hurdles for incident response and compliance monitoring.
Security Incidents and Agentic Risks
The risks associated with autonomous AI are already materialising in the corporate sector. Dashlane found that 72% of IT and security executives reported their organisation had experienced a security incident involving an AI agent—either through compromised credentials or unauthorised access—within the past 12 months.
Beyond credential sharing, the study noted that 23% of UK workers have copied sensitive information into AI prompts that they would not normally share with a colleague or a standard website. This behaviour increases the risk of data leakage, as information shared with large language models (LLMs) can sometimes be stored on third-party servers or used for model training, potentially leading to the exposure of proprietary data.
Governance and Mitigation
The rapid adoption of agentic AI has prompted national regulators to issue warnings. The National Cyber Security Centre (NCSC) has previously published guidance on securing AI agents, urging organisations to apply the principle of least privilege and robust authentication to non-human identities.
Security leaders currently face a difficult balance, with 68% of UK IT executives reporting pressure to deploy AI capabilities quickly while simultaneously maintaining an acceptable security posture. Experts suggest that the fundamental principles of identity and access management (IAM)—such as just-in-time provisioning and strict credential hygiene—must now be extended to AI “colleagues” to mitigate the risk of hijacking or data exfiltration by threat actors.



