A sales rep asks an agent to prepare account notes before a call. An engineer runs a coding agent that opens pull requests overnight. Finance has an agent that reconciles invoices on a schedule, and nobody starts it by hand.
That's the modern workforce: people, the personal agents that act for them, and autonomous agents that run on their own with a person accountable for them. At NewCore, this is the workforce we secure.
Identity was built for the first group. Sign-in, MFA, lifecycle, and access reviews all assume a person sits behind every session. Agents break that assumption, and the accountability questions they raise are landing on security teams now. This post looks at what that means for protecting identity, today and as the workforce keeps changing.
The problems teams have today
A risky person makes their agents risky. If someone's account is phished, every agent acting on their authority is now acting for an attacker too. Most stacks see the person's risk in one place and the agent's activity in another, so the two rarely get connected in time.
Personal agents act on their owner's authority, so a valid credential can't tell you whether the owner meant a given action. Take a hypothetical sales manager who connects a personal agent to their inbox and CRM to tidy up the pipeline. One customer email carries hidden instructions, and the agent follows them: it exports the account list and sends it to an outside address. Every step used the manager's own access, and every check passed. The manager never intended any of it.
Autonomous agents outlive their sponsors. The person who set one up changes teams or leaves, and the agent keeps running with the same access. Nobody is left who knows what it was for.
Approvals swing between two bad options. Ask the owner about everything and they stop reading the prompts. Ask about nothing and approval turns into a rubber stamp. Either way, the owner stops being a real check.
The market has answers for parts of this. There are good tools for the device, for cloud access, for finding agents nobody registered, for detecting identity threats, and for signing people in. Each one covers a link. The gaps sit where one tool hands off to the next, and none of them is the source of identity for the whole workforce.
Where the workforce is going
Agents are getting more autonomous. They run longer tasks, call other agents, and make more decisions without a person watching each step. The chain from a person to the action that finally happens keeps getting longer.
The workforce is also getting a physical side. GXO announced it will expand Agility Robotics' Digit humanoids to more U.S. warehouses starting in early 2027, and Figure reported that its humanoids worked 10-hour shifts on BMW's Spartanburg line for 11 months.
A humanoid on a warehouse floor is an autonomous agent with a body, and someone is accountable for it. The identity questions stay the same as for software agents, with higher stakes, because a wrong action moves something physical. Identity for physical agents is still an open question for the whole industry, and I expect the answer to start from the same place: a person behind every agent.
How NewCore is built for the modern workforce
NewCore is one IdP designed for the whole modern workforce, with people, personal agents, and autonomous agents in a single model. Most identity platforms started with people and extended to agents through add-ons, which makes sense given where they started. We designed NewCore around the workforce as it looks now, with an accountable person behind every agent. The same should hold when agents create agents. An agent created by another agent should be owned by whoever was accountable for its creator, so the chain of accountability still leads back to a person.
When a person's account turns risky, every agent acting for them should be treated as risky too. Risk should flow the other way as well: an agent that starts behaving badly should raise its owner's risk. Every risk call should show the reasons behind it, and when someone leaves, the agents acting for them shouldn't quietly keep going.
The owner also needs a better way to stay involved. Each agent action should be checked against the intent behind it, at the moment it happens. The human owner should be asked only when an action really needs it. Owners should hear about real risk or a real anomaly, and not much else.
Both depend on knowing who stands behind an agent, and the IdP is where that link is created.
Where to start
- Name an accountable person for every agent in production, including agents that other agents created.
- Decide what happens to each agent when that person changes teams or leaves, before it happens.
- Connect each person's risk signals to the agents acting for them, so a compromised account doesn't leave its agents running unnoticed.
- Check how often owners are asked to approve agent actions. If they approve almost everything, the prompts have stopped being a check.
Takeaway
Protecting identity now has to cover the whole chain, not just the login. The chain is the person, their personal agents, their autonomous agents, and the agents those agents call. It only gets longer as agents gain autonomy and, increasingly, a body.
What does that chain look like in your company today, and where do you lose track of who an action was really for?