AI phishing hits 58% despite high detection: Yubico

Phishing detection reached 59%, above the 53% global average.

Nearly three in five Singapore respondents reported at least one successful artificial intelligence (AI)-driven phishing attack within their organisation over the past 12 months, the highest amongst the surveyed markets.

Yubico’s annual Global State of Authentication Survey found that 58% of Singapore respondents reported a successful AI-driven phishing attack, compared with 44% globally.

The findings come as 81% said their organisations enforce multi-factor authentication across all applications and services.

Singapore respondents were also the most successful amongst surveyed markets at identifying an AI-generated phishing email, with 59% doing so correctly compared with 53% globally.

Phishing attempts have continued to rise, with 78% reporting an increase in attacks targeting their companies over the past year.

Seventy-two percent said their organisations use different forms of authentication across applications, with mobile TOTP or push-based authenticator apps the most commonly used method at 49%.

Usernames and passwords were also widely used by 41% of respondents, alongside other methods including mobile SMS-based authentication, device-bound passkeys, hardware security keys, and synced passkeys.

“Rather than replacing passwords, stronger authentication methods are increasingly operating alongside older, phishable ones, adding friction for employees navigating different login requirements,” Yubico said.

Yubico added that 58% of respondents were issued a traditional username and password when they joined their organisations.

Meanwhile, AI agents are introducing a new identity-security challenge as they take on more tasks on behalf of employees.

Almost all respondents (98%) said verifying the identity and authenticity of AI agents is important, whilst 95% said it is important to review and approve actions before an AI agent executes them on their behalf.

However, 51% are comfortable allowing AI agents to make low-risk operational decisions, whilst 60% are comfortable allowing them to communicate with colleagues or clients on their behalf.