Retiring twenty years of long-lived passwords and access keys does not happen at a product launch. It happens on the day a feature ships to every customer, and only one of the two BeyondTrust features built to do the retiring has a date on it. NHI Governance, four months out. Workload Credentials, its counterpart, does not have one yet.

BeyondTrust used Black Hat USA 2026 to announce the first four features on its Pathfinder platform, an attempt to track every account, whether it belongs to a person or a piece of software, and control what that account can do (BeyondTrust, 2026). The announcement bundles four pieces: PathfinderAI and a new Model Context Protocol, or MCP, connection, AI Agent Security, NHI Governance for accounts that belong to software and AI agents rather than people, and a newly announced Workload Credentials feature (BeyondTrust, 2026).

PathfinderAI lets security teams ask plain-English questions about who has access to what, instead of digging through spreadsheets and logs. The MCP connection opens that same information to outside AI tools. Microsoft Copilot, OpenAI's models, and Claude can now check BeyondTrust's access records through one shared connection method, instead of BeyondTrust building a custom hookup for each tool (BeyondTrust, 2026).

AI Agent Security controls what those same tools are allowed to do once they are running on an employee's computer.

It sets rules for tools like Claude Code, Copilot, Cursor, and OpenAI's Codex before they take action, and it flags AI tools a security team never approved or even knew about, a pattern most vendors now call shadow AI (BeyondTrust, 2026). Chief Technology Officer Marc Maiffret frames the whole bundle as identity strategy extended to a new class of actor, not artificial intelligence bolted onto an existing product line (BeyondTrust, 2026).

The Credential Problem Is Not Theoretical

BeyondTrust's own research team, Phantom Labs, supplies the announcement's urgency. The team ran more than 400 projects testing how attackers break into company systems, and in most cases the weak point was who had access to what, not a software bug. Amazon Web Services, Microsoft's identity system Entra ID, GitHub, Okta, and Salesforce showed up most often as the places where that weak point lived (BeyondTrust, 2026).

One project found a specific problem inside AWS Bedrock, Amazon's AI development service. Teams often set up long-term digital keys, the kind meant to work indefinitely, generate them once, and forget about them. Those keys can quietly create a hidden account with broader permissions than anyone intended, and the key itself can leak information an attacker uses to find and target the company's cloud account (BeyondTrust, 2026). Short-term keys work differently. They borrow the permissions of whoever is already logged in, expire in a matter of hours, up to twelve, and never create that hidden account, which is why BeyondTrust wants the long-lived kind retired. The company released a free tool, bedrock-keys-security, so customers can find this exposure themselves instead of waiting to get breached.

Phantom Labs separately measured AI agents inside client companies growing 466.7 percent year over year, based on BeyondTrust's own usage data. That kind of growth turns a housekeeping problem into a scale problem (BeyondTrust, 2026).

Visibility Became the Easy Part

Most vendors selling into this category can hand a chief information security officer, or CISO, a dashboard now. Visibility used to be the differentiator. It has become table stakes, which pushes the real competition into what a platform lets a CISO do once the dashboard stops being new.

Palo Alto Networks made a related argument in May, folding its CyberArk acquisition into a platform called Idira and citing research showing machine identities now outnumber human ones 109 to 1, with 79 of those 109 being AI agents (Bellamkonda, 2026). Delinea took the opposite entry point in July, shipping a feature that checks and approves what an AI agent is trying to do at the moment it tries to do it, rather than keeping a list of which agents exist (Delinea, 2026).

BeyondTrust's four new pieces sit between those two approaches. PathfinderAI and the MCP connection compete on the same visibility ground Idira and Delinea's own monitoring already cover. AI Agent Security and NHI Governance compete on control, the piece Delinea staked its bet on and where BeyondTrust brings two decades of access-management history that neither newer competitor has.

Unknowns and Uncertainties

All four modules ship through early access rather than general release. NHI Governance carries the only public general availability date, planned for the fall in the United States, with other regions to follow (BeyondTrust, 2026). Workload Credentials, the newest of the four, has no early access date yet; BeyondTrust says access is planned to open soon.

What Happens Next

Three questions worth tracking. Does the fall date for NHI Governance hold, or does the industry keep announcing new features faster than it can ship them? Felix Gaehtgens spent thirteen years as an outside analyst studying this exact market, and now he sets BeyondTrust's product strategy from the inside; does that make the roadmap the model other vendors get measured against, or an informed bet? And with Palo Alto Networks' Idira and Delinea's new AI-agent controls already live before BeyondTrust's own version is available to everyone, does two decades in this field outweigh whichever company finishes building AI account controls first?

About Shashi Bellamkonda

Loading bio…

LinkedIn Profile

Get new posts by email:

Subscribe

Powered by