The Security Risk Manish Meshram Sees Growing Alongside Life Sciences Automation

Reese Watson - Author
By

Published Nov. 17 2026, 1:15 p.m. ET

Manish Meshram
Source: Gabrialle Smith

As healthcare and life sciences systems adopt more automation and AI, Manish Meshram is focusing on a less visible risk: the identities used by machines, applications, and automated processes to access regulated enterprise systems.

Article continues below advertisement

As artificial intelligence moves deeper into healthcare and life sciences operations, Manish Meshram is paying attention to a security problem that receives less attention than the models themselves. The SAP solution architect has been working with senior security professionals and conducting workshops around nonhuman identity theft, access, and governance in regulated systems. His concern is straightforward: every automated process still needs a way to connect to enterprise data, and those connections have to be governed within the rules surrounding the systems they access.

In one recent initiative for a large pharmaceutical company, Meshram has been directly involved in defining and implementing security protocols and standard operating procedures for AI agents and other nonhuman identities. His work includes policy enforcement, policy decision-making, and fine-grained authorization intended to protect sensitive personal and clinical information, unpublished intellectual property such as compound structures and formulations, and commercially sensitive trial results. The objective is to ensure that AI agents operate with least privilege, auditable, policy-driven access instead of inheriting broad permissions that may expose regulated or confidential information.

Article continues below advertisement

Meshram’s focus grows out of the way enterprise systems are changing. As AI becomes more closely connected to enterprise data and applications, organizations have to account for the identities and permissions involved in that access. In healthcare and life sciences, Meshram says those identities need clear permissions and governance because they operate inside regulated environments.

“Giving an automated system access does not remove the security responsibility,” Meshram said. “You still have to know what identity is being used, what it is allowed to access, and how that access is controlled.”

The phrase “nonhuman identity” can refer to the digital identities software uses to authenticate and gain access to systems. These can include service accounts, API keys, OAuth clients, certificates, tokens, bots, RPA accounts, and CI/CD pipeline credentials. AI agents represent a newer and more powerful category because they may choose which tools to call and which data to access instead of following only a fixed script.

Article continues below advertisement

Crave InfoTech is also developing products around the same governance problem. AccessHub.AI is designed to govern human, machine, and AI identities across SAP environments, including access risk and transaction traceability. Its focus reflects the broader enterprise shift toward governing automated identities as they gain access to more systems and data.

Meshram says GxP, 21 CFR Part 11, and EU Annex 11 make traceability especially important in regulated systems, including audit trails that establish who or what performed an action and when it occurred. His approach is to give agents unique and traceable identities and, where possible, capture the human or process on whose behalf the agent was acting. He also distinguishes between agents acting under their own identity and agents operating through delegated user access, because an agent should not be able to exceed the permissions of the person requesting the action.

Article continues below advertisement

For Meshram, the important question is whether the identity behind a connection has appropriate access and whether the organization has a defined way to control it. In regulated systems, permissions around automated identities become part of the broader governance responsibilities surrounding the applications they use.

“Security cannot be added after the process is already designed,” Meshram said. “You need to understand what the automated system needs to do and then give it the access required for that purpose, without opening more of the environment than necessary.”

In the pharmaceutical initiative, Meshram introduced a centralized policy decision and enforcement model for AI agent access. Before the change, agents could rely on static service accounts or broad role-based permissions, while authorization logic was handled separately inside individual applications. Meshram says that structure made access inconsistent and difficult to audit.

Article continues below advertisement

He defined a central policy decision point that evaluates agent requests based on the agent’s identity, the requested action, the sensitivity of the data, and the entitlements of the user the agent is acting for. Policy enforcement points at the tool, API, and data layers then carry out those decisions. Meshram also moved the environment toward fine-grained, default-deny authorization so an agent cannot exceed the requesting user’s access, and high-impact actions require human approval. He wrote SOPs covering agent registration, ownership, access reviews, recertification, and decommissioning.

Electronic signatures and approvals create an especially important boundary. Meshram’s approach keeps a human identity involved at signature and release points instead of allowing an AI agent to perform those actions independently. That preserves meaningful authentication and keeps consequential approvals connected to an accountable person.

Article continues below advertisement

The initial implementation produced several measurable changes. Meshram reports that seven AI agent identities were onboarded under the centralized policy model during the first nine months. About 45 shared service accounts and static API keys were retired or replaced with short-lived, vaulted credentials, while roughly 30 percent of standing permissions were removed during initial access reviews. More than 60 fine-grained authorization policies were approved for data classes including protected health information, unpublished compound and formulation intellectual property, and trial information.

Meshram also reports that agent actions involving those regulated data classes now generate audit records capturing both the agent and the human on whose behalf it acted. The SOPs for agent registration, ownership, access recertification, and decommissioning were approved by security, privacy, and quality teams and adopted as the standard for new agent deployments. Those controls make access decisions more consistent and give organizations a clearer record of how nonhuman identities interact with sensitive systems.

Article continues below advertisement

Identity governance becomes more consequential when automated processes interact with applications supporting regulated activity. Traditional enterprise applications often rely on form-level or screen-level authorization, while AI agents may interact through APIs and other protocols. Meshram has focused on designing more granular policies for environments where multiple agents can interact with enterprise systems and with one another.

In life sciences, automated identities can interact with applications supporting business-critical processes, which makes access decisions part of a wider governance responsibility. Meshram’s work treats identity as one of the controls that has to be defined as automation expands.

For regulated healthcare and life sciences organizations, that creates a familiar enterprise responsibility in a new form. Systems still need credentials, permissions, auditability, and controls even when software initiates the activity. Meshram’s current work focuses on building that governance layer into automated identities before broader adoption allows access to grow faster than the controls designed to contain it.

Advertisement

Latest Business News News and Updates

    © Copyright 2026 Engrost, Inc. Distractify is a registered trademark. All Rights Reserved. People may receive compensation for some links to products and services on this website. Offers may be subject to change without notice.