AI Agents Are Borrowing Their Creator's Credentials
By @sharedot · · 6 pages
- AI Frontier
- Agent Security
- Credentials
- Oauth
Nudge Security finds most employee-built AI agents run on their creator's OAuth grants or API keys, leaving logs and identity providers blind.
The agent that is secretly you
Help Net Security reports that AI agents employees build inside project trackers, CRMs and automation tools often run on the credentials of the person who created them, carrying that person's access to company data.
Among organizations with at least one AI agent, the median organization had seven, about 44% had more than 10, and the agents ran across 15 different platforms — with over half of organizations running agents on two or more.
Common homes include Atlassian's Rovo Agents, Zapier Agents, Claude Managed Agents, Cursor Automations and ChatGPT Agents. Some platforms give an agent a dedicated bot account, but often the agent works off the OAuth grant or API key of whoever set it up, or off a connection a team shares.
Why security teams can't see it
The surprise is architectural: these agents run inside the platforms themselves, off company devices, so endpoint tools miss them — and when they have no identity of their own, the identity provider has nothing new to manage.
Help Net Security quotes Nudge Security's finding that logs show the employee, not the agent, making it hard to tell which actions a person took and which their agent took, and that an agent's access may stay the same when its creator changes roles.
CTO Jaime Blasco told Help Net Security: "The project tracker, the CRM, the chat tool: each one now lets an employee build something that acts on their behalf, with their access, continuously... The question for security teams is no longer whether that's allowed. It's whether they know it's happening."
OAuth grants are the wider blind spot
The same report, as covered by Help Net Security, found an Anthropic integration in 74% of organizations and an OpenAI integration in 67%, with Google Workspace the most common endpoint for both.
Most links are OAuth grants created when an employee clicks "allow" — data then travels straight between cloud services and never crosses the corporate network, out of sight of AI gateways and data loss prevention tools. Those grants persist until revoked, often long after use stops.
Help Net Security also recalls the 2025 Salesloft Drift breach, in which attackers stole OAuth tokens customers had granted to Drift's AI chat tool and used them to pull data from hundreds of companies' Salesforce environments.
Pasted content or uploaded files appeared in 18% of observed prompt activity.
Liability shifts to whoever hands over the keys
The credential problem now reaches consumers too.
TechCrunch reports that Citibank, the third-largest US commercial bank, updated its Online User Agreement to say customers who grant account access to third-party AI agents are responsible for activities conducted through those agents — including, per TechCrunch, anything hackers do if the AI tool leaks their credentials or is hacked.
JPMorgan and Bank of America announced similar policies last month, PNC changed its terms the same week, and Citizens emailed customers a similar warning, according to TechCrunch.
Anthropic's response is telling: TechCrunch notes its new personal finance tool is explicitly "read-only" and "can't move, manage, or invest your money."
Nudge Security's advice to organizations: inventory agents inside business platforms, learn whose access each uses, and revoke OAuth grants nobody needs. Blasco expects the gap between what identity providers know and what acts on company data to keep widening.