Published 7 November 2026
Microsoft: AI agents found exposed secrets and reached an Azure vault in five minutes
Attackers used an AI agent to find exposed credentials on GitHub, then reach an organisation’s Azure Key Vault. Microsoft says the sequence took just five minutes.
That is the sharpest detail in Microsoft’s 5 November warning about threat actors using agentic AI to search public code repositories for exposed secrets and act on what they find. The secrets included application credentials and API keys. Microsoft says it observed this activity from August 2026. Microsoft Security Blog
The risk is not that AI has made attackers unstoppable. It is that it can help them find and exploit familiar security mistakes faster, with less hands-on effort. For businesses, that can leave less time to spot exposed credentials, revoke them and contain the damage.
What Microsoft observed
Microsoft says threat actors used an agentic offensive framework called ARTEX to search public GitHub repositories for secrets. The agents could analyse what they found and carry out follow-on actions, reducing the need for an operator to guide every step.
In one case, an agent moved from searching GitHub for secrets to exfiltrating data from Azure Key Vault in five minutes. Microsoft also described later activity in which attackers used valid application credentials to access cloud resources and seek out further credentials and data. Microsoft Security Blog
This is a specific Microsoft-observed campaign, not proof that every exposed secret will be found or abused by an AI agent. But it shows how quickly a leaked credential can become a path into connected systems.
Why it matters
Secrets are the credentials that let applications and services access systems: API keys, tokens, passwords, connection strings and application credentials. They can end up in public code, configuration files or other places where they should never have been stored.
A secret that is exposed may remain usable even after the file or post containing it has been deleted. Attackers may already have copied it, and removing the original exposure does not revoke the credential. Microsoft advises organisations to treat exposed secrets as compromised, rotate or revoke them, and investigate how they were used. Microsoft Security Blog
The broader concern is straightforward: organisations are connecting more applications, cloud services and AI tools, and those connections rely on credentials. The more widely a secret is shared, and the longer it remains valid, the more useful it can be to an attacker.
I have written before about how AI is changing the cyber risk conversation. In CrowdStrike’s AI Agent Kill Switch Is a Sign of the Times, I looked at the need for controls over autonomous agents. In Australia Needs an AI Early Warning System. This Week Proved Why, I explored how quickly abuse can move. Microsoft’s latest findings bring those concerns back to a basic security problem: credentials that are exposed and still work.
Phil’s analysis
My view is that this is an operations problem with an AI accelerator. The underlying weakness is familiar: credentials are left in code, old tokens stay active, access is broader than necessary, or nobody is sure who owns a service account.
AI does not create those weaknesses. It can help attackers search for them at scale, assess what they find and move on to the next step faster. That makes routine security hygiene more urgent.
It also raises practical questions for organisations deploying AI agents: what credentials can an agent access, what can it do with them, how is that access logged, and how quickly can it be revoked? These are essential questions whenever an agent can interact with business systems or data.
The lesson is not that every organisation needs a new AI security product. It is that secrets need to be managed as live access, with clear ownership, limited permissions and a reliable process for rotation and revocation.
What to do now
If you manage business systems, cloud services or AI tools:
- Find your secrets. Check code repositories, configuration files, deployment pipelines and shared storage for API keys, tokens, passwords and service credentials.
- Revoke or rotate anything exposed. Deleting a secret from a repository does not make it safe again.
- Reduce access. Give each person, application and service only the permissions it needs.
- Shorten credential lifetimes. Use managed identities or short-lived credentials where they are supported.
- Review activity. Check for unexpected sign-ins, secret reads, downloads or changes to application permissions.
- Include AI agents in access reviews. Know which credentials they can use, what actions those credentials permit, and how to disable access quickly.
Microsoft’s warning is a reminder that forgotten credentials can provide a direct route into cloud systems. With agentic tools speeding up the search and follow-on activity, the time to find and fix exposed secrets is now.
Sources
- Microsoft, Threat actors use agentic AI tooling to find and abuse exposed secrets, 5 November 2026. Microsoft Security Blog
How this analysis is researched, sourced and corrected: Editorial standards and corrections. Spotted an error? Contact Philip Hall.
