On October 16, 2026, Straiker, a company that styles itself "The Agentic Security Company," announced an integration with Workday timed to the general availability of Workday's Agent Passport: it wants to connect its Ascend AI red-teaming platform to the trust layer of Workday's Agent System of Record, so that AI agents built by Workday customers and partners can carry verifiable evidence of how they were tested, by whom, and against which Workday-defined criteria, before anyone installs them.
The headline is straightforward: Straiker says it is the first independent security partner to red-team agents built by Workday customers and partners for the Workday Marketplace. The design principle underneath is the part worth stealing, because it applies to every agent deployment, on every platform, including yours.
Tested by one party, vouched for by another
Agent Passport splits the trust problem three ways. Workday defines the claims an agent should be able to satisfy. Independent partners, Straiker being the first, test agents against those claims. Workday alone issues the resulting Stamps. As the announcement puts it, the party that tests an agent is never the party that vouches for it.
The quotes make the intent unusually clear. "Workday has set a standards-based bar for what a trusted agent looks like, and Agent Passport gives that bar teeth," said James Beasley, vice president of business development at Straiker. And from the platform side, Dean Arnold, VP of Workday's AI Platform: "As AI agents take on real work in HR and finance, 'trust us' is not a governance strategy. Customers need to know exactly what has been verified about an agent, by whom, and to what standard."
What Ascend AI actually probes for is the standard adversarial catalogue: prompt injection, tool misuse, data leakage and other attack paths drawn from the OWASP LLM Top 10, the OWASP Agentic Security Initiative and MITRE ATLAS. Agents that handle HR, payroll and financial data carry a different risk profile than most, the release notes, because a wrong answer in the people-and-money domains Workday powers can lead to a regulatory event or an audit finding.
What a Stamp proves, and what it doesn't
A Stamp records three things: the risk that was tested, who tested it, and which version of the agent the test applies to. The third item is the quiet breakthrough. When the agent changes, the Stamp does not follow it. Evidence that cannot outlive the software it describes is the difference between a trust program and a rubber stamp.
Adversarial Resilience, Straiker's category, joins six existing Trust Categories in the Passport framework: Build Integrity, Runtime Safety, Audit Integrity, Data Sovereignty, Delegated Authority & Human Control, and AI Governance. In plain language, together they cover how an agent was built, how it behaves live, what its logs prove, where its data lives, who can override it, and whether anyone is governing at all. Each category draws on a different independent source, and each customer decides which Stamps its agents must hold before activation.
| Question a customer asks | Passport's answer | Why it beats a vendor questionnaire |
|---|---|---|
| Has anyone attacked this agent? | Adversarial Resilience Stamps from independent testers | Testers don't profit from a pass |
| Does the evidence match this version? | Stamps are cryptographically tied to the version tested | Evidence expires when the code changes |
| Who says so? | A named, independent testing party | Attestation becomes attribution |
| Who decides what "good" means? | Workday-defined criteria per Trust Category | Claims are public, testable, comparable |
| Who enforces it? | The platform's system of record blocks unstamped installs | Procurement teeth, not policy prose |
Two caveats belong in the fine print. Straiker's integration is "targeted for Agent Passport's general availability" and still must complete Workday's partner technical review, which the release itself flags in standard unreleased-features language. And the first-party framing matters: this is a vendor announcing a vendor partnership, so treat "first independent security partner" as Straiker's claim, reported October 16, 2026, rather than an audited fact about the whole market.
Why this lands the same week as everything else
Agent Passport's timing is not accidental. Two weeks ago Workday closed its acquisition of C1.ai to give every employee and AI agent one governed identity, a story we covered in depth. ServiceNow announced its own AI Passport on October 15, 2026: crypto-signed credentials verified at every step, which we also covered. India's CERT-In put out a critical advisory on October 15 about agents reaching production systems undetected, and OpenAI spent this week adding four more misalignment cases to its incident disclosure page. SailPoint's numbers from earlier this month, 79% of organizations running agents in production, 2% securing them, frame the whole sprint.
The pattern across every one of those stories: the industry has stopped arguing about whether agents need governance and started shipping competing versions of what governance looks like. Credential-based identity (ServiceNow, Workday's C1), independent adversarial testing (Straiker, and platforms like Hack The Box's AI Range we covered earlier), disclosure regimes (OpenAI, the EU AI Office): these are the same bet placed at different layers.
The pattern is the product: four steps to steal
If you never touch Workday, the architecture still copies cleanly into an organization of any size:
- Define testable claims per agent. Not "the agent is secure," but concrete, falsifiable statements: which scopes it may touch, which actions it must refuse, what it must never send outside the boundary.
- Commission testing from someone who does not profit from the result. Internal red-teaming is a start; independent testing, by another team or an external firm, is the actual bar. If the tester and the vouching party are the same person, you have a questionnaire, not evidence.
- Pin every result to a version. Re-test after any change, and make evidence visibly expire. The Stamp model works precisely because "we tested it once, in June" stops counting the moment the code moves.
- Put the evidence where the install decision happens. Results buried in a vendor binder or a shared drive get skimmed; results attached to the system of record, blocking activation until Stamps are present, get enforced. Even a lightweight internal registry, agent name, version, claims tested, tester, date, does the job.
Conclusion
Agent Passport's most exportable idea is not the cryptography; it is the separation. Workday defines what a trusted agent means, independent parties try to break the agent against those claims, and only the platform, which cannot also be the tester, issues the signed record pinned to the exact version that was tested. That triangle, public test claims, testing by someone who does not profit from a pass, evidence that expires the moment the agent changes, is what turns agent governance from slideware into an operating control. The October governance sprint, from ServiceNow's AI Passport to the EU's first enforcement actions, keeps arriving at the same conclusion from different directions: the market no longer accepts "trust us" as an agent security posture. If you want help turning that principle into a procurement checklist for the agents entering your organization, talk to us.
Read next:
