Security researchers at VulnCheck say an internet address that Microsoft previously tied to an AI agent designed to hunt for leaked secrets on GitHub has been used to break into Langflow, IBM's tool for building AI agents, and that the intruder has developed new ways to gain entry.
Security alert · Langflow intrusion
An address tied to a secret-hunting AI agent is now linked to attacks on Langflow
Researchers say the address, previously connected to an AI agent that scans GitHub for leaked credentials, was used to break into Langflow and that the intruder developed new ways to gain entry.
CVE-2025-3248
The flaw used to access Langflow
May 2025
Listed on CISA's Known Exploited Vulnerabilities list
Unconfirmed
Number of affected organizations and scope of data taken
How the reported attack chain works
Automated tooling links exposed credentials to the systems they may unlock
STEP 1
Credentials leak
Secrets exposed in public code repositories
→
STEP 2
Agent hunts
A GitHub-focused AI agent finds the secrets
→
STEP 3
Access sought
Linked address pursues the systems behind the credentials
→
STEP 4
Langflow breached
Exploit of CVE-2025-3248 gives entry, then further steps
Recommended checks for security teams
Confirm Langflow deployments are patched against CVE-2025-3248
Review and rotate any credentials that may have been exposed
Monitor for activity tied to known malicious addresses
What remains unclear
How many organizations were affected
What data was compromised
Whether the earlier attribution covers all observed activity
Key takeaway
Known vulnerabilities in AI development tools can be weaponized quickly once they are publicly documented.
As teams build workflows on AI agent frameworks that connect to models, APIs and internal data, one compromised platform can expose a wide range of sensitive assets.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...