50%
of security leaders identify attacks targeting AI systems as one of their biggest preparedness gaps.
Source: PwC 2027 Global Digital Trust Insights. Independent; 3,934 business and technology leaders, 2026
Public sources only
Why now.
AI agents are reaching live systems with credentials nobody meant them to use. The signal is now in advisories and surveys, not just anecdotes.
In the news, October 2026
CERT-In: AI agents as attack vectors.
On 15 October 2026, India’s CERT-In issued an advisory rated Critical (CIAD-2026-0049, “AI Agents as Attack Vectors”). It says AI model providers disclosed that their models, during research and evaluation, “had reached the live systems of Indian organizations,” reading business records “through credentials and connection settings exposed in application configuration, public code repositories and permissive cloud service roles.”
“None of the affected organizations detected the access.”
The root causes were ordinary. Among the advisory’s recommendations: audit repositories “(including commit history)” for exposed credentials, rotate them, and give each AI agent run “a separate, short-lived identity with least privilege.”
Source: CERT-In advisory CIAD-2026-0049, 15 October 2026. Press coverage: Mint / PTI, 16 October 2026.
What surveys report
The gap, in other people’s numbers.
22%
would authorise fully autonomous AI actions.
Source: PwC 2027 Global Digital Trust Insights. Independent; 3,934 business and technology leaders, 2026
Figures are quoted as published by the source. We link every source so you can check it.
One question to add.
For each agent you run, which of those credentials could it actually use? That’s where our Agent Governance Readiness Assessment is designed to start.