---
title: "Palantir's $4.9M IRS AI Agent Order, Read Line by Line"
slug: "palantir-irs-ai-correspondence-agent-contract"
author: "Miguel Amigot"
date: "2026-11-19 12:00:00"
category: "Premium"
sector: "government"
topics: "Palantir IRS contract, Treasury AI contract, IRS AI correspondence agent, Schedule C, federal AI procurement, BPA call order, OMB M-25-22, vendor lock-in, government AI agents, AI code ownership, sovereign AI"
summary: "The reported $5M Treasury AI contract is a $4.9M IRS order to Palantir for an AI correspondence agent for Schedule C, placed on a Palantir-only purchase agreement. What the public record shows, what it does not, and the ownership questions an agency should settle before ordering an AI agent."
banner: ""
thumbnail: ""
linkedin: |
  "Palantir won a $5M AI contract with the US Treasury." The public record is more specific, and more useful.

  On November 5, 2026 the IRS signed a $4,899,521.50 order with Palantir to design, develop and deploy an AI correspondence agent for Schedule C, with agentic AI capabilities, workflow integrations and "semantic ontology governance". It runs four months, November 6, 2026 to March 5, 2027.

  It is not a standalone contract. It is a call on a single-award IRS purchase agreement under the IRS Compliance Hub program, signed September 29, 2026, running to 2030, with a potential value of $250 million.

  Five Palantir orders starting between September 23 and November 6 total about $53.1 million.

  The question for every agency is not who won. It is who owns the agent afterwards: its code, its ontology, its prompts and the correspondence it handles. OMB's M-25-22 asks for rights to code produced under a contract. A licensed platform underneath is a separate negotiation.

  With ibl.ai you own all the code and the data, run any model, and keep the agent when the order ends.

  #iblai #AgenticAI #EnterpriseAI #GovTech #FederalAI #AIProcurement
faqs:
  - question: "Did Palantir win a $5 million AI contract with the US Treasury?"
    answer: "The Treasury bureau involved is the IRS. On November 5, 2026 it signed a $4,899,521.50 order with Palantir for an AI correspondence agent for Schedule C, running November 6, 2026 to March 5, 2027. USAspending lists it as a call on an IRS blanket purchase agreement, not a standalone contract."
  - question: "Is the $4.9 million a ceiling or money obligated?"
    answer: "USAspending reports $4,899,521.50 obligated, equal to the base and exercised options, and shows no larger options value for the order. The parent purchase agreement shows no obligations of its own but a potential value of $250 million; spending happens through calls like this one."
  - question: "Who owns the code of an AI agent an agency orders?"
    answer: "That depends on the order's terms, which the public spending record does not show. OMB memo M-25-22 asks agencies to secure rights to code and models produced in performance of a contract, but a commercial platform licensed underneath the agent is a separate licensing question the agency has to raise."
---

## The Short Answer

**The "$5 million Treasury AI contract" is a $4,899,521.50 IRS order to Palantir, signed November 5, 2026, to build an AI correspondence agent for IRS Schedule C over four months. It is a call on a Palantir-only IRS purchase agreement, not a new contract. The question it raises applies to every agency buying agents: on ibl.ai you own all the code and the data.**

The headline came from a social post. This piece reads the federal spending record instead, because the record answers what the headline cannot: what was bought, how, for how long, and what the agency will hold when the order ends.

## What did the IRS actually buy from Palantir in November 2026?

The IRS bought an AI agent to handle correspondence for Schedule C, the form on which, [per the IRS](https://www.irs.gov/forms-pubs/about-schedule-c-form-1040), a sole proprietor reports income or loss from a business they operated.

The [USAspending award record](https://www.usaspending.gov/award/CONT_AWD_205AE927F00003_2050_205AE926A00004_2050) describes the work as the design, development and deployment of AI-based assistants, agentic AI capabilities, workflow integrations and semantic ontology governance "to build an AI correspondence agent for the IRS Schedule C."

The order is 205AE927F00003. The IRS IT Operations office both awarded and funded it, and the record files it under the IRS major program "Compliance Hub" with product code DA10, application development software as a service.

<table style="width:100%; border-collapse:collapse; margin:1.5rem 0; font-size:0.95rem;">
  <thead>
    <tr style="background:#f5f5f0; border-bottom:2px solid #2175C5;">
      <th style="text-align:left; padding:0.75rem; color:#5f6368;">Field in the federal record</th>
      <th style="text-align:left; padding:0.75rem; color:#5f6368;">Palantir IRS order 205AE927F00003</th>
    </tr>
  </thead>
  <tbody>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Awarding bureau</strong></td>
      <td style="padding:0.75rem;">Internal Revenue Service (Department of the Treasury), IT Operations</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Signed</strong></td>
      <td style="padding:0.75rem;">November 5, 2026</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Period of performance</strong></td>
      <td style="padding:0.75rem;">November 6, 2026 to March 5, 2027</td>
    </tr>
    <tr style="background:#f0f9ff; border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Amount obligated</strong></td>
      <td style="padding:0.75rem; font-variant-numeric:tabular-nums;">$4,899,521.50</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Award type</strong></td>
      <td style="padding:0.75rem;">BPA call on IRS BPA 205AE926A00004</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Parent vehicle of the BPA</strong></td>
      <td style="padding:0.75rem;">GSA Multiple Award Schedule contract 47QTCA24D004L</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Work</strong></td>
      <td style="padding:0.75rem;">AI correspondence agent for IRS Schedule C</td>
    </tr>
  </tbody>
</table>

## Is the Palantir IRS award a $5 million contract or a ceiling?

The Palantir IRS figure is money obligated, not a ceiling, and it is not a standalone contract. USAspending reports $4,899,521.50 obligated, equal to the base and exercised options, and lists no larger options value for the order.

It is a call on [IRS blanket purchase agreement 205AE926A00004](https://www.usaspending.gov/award/CONT_IDV_205AE926A00004_2050), described as the "Compliance Hub - Compliance Vertical Federal Platform" BPA. That agreement was signed September 29, 2026 and runs to September 29, 2030.

The BPA is recorded as a single award, with one offer received, placed against Palantir's GSA Multiple Award Schedule contract.

The BPA itself shows $0 obligated and a potential value of $250 million, including all options. Money moves only through calls like this one, so the agreement is the channel and each call is a purchase.

## How does the Schedule C agent fit Palantir's other IRS work in 2026?

The Schedule C agent is the latest of five Palantir orders the IRS placed in about six weeks. A USAspending search of Palantir's Treasury awards lists them, and together they come to about $53.1 million obligated.

<table style="width:100%; border-collapse:collapse; margin:1.5rem 0; font-size:0.95rem;">
  <thead>
    <tr style="background:#f5f5f0; border-bottom:2px solid #2175C5;">
      <th style="text-align:left; padding:0.75rem; color:#5f6368;">Start date</th>
      <th style="text-align:left; padding:0.75rem; color:#5f6368;">IRS order (as described on USAspending)</th>
      <th style="text-align:right; padding:0.75rem; color:#5f6368;">Obligated</th>
    </tr>
  </thead>
  <tbody>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;">Sep 23, 2026</td>
      <td style="padding:0.75rem;">Compliance Hub IRS case selection: software licenses plus operations and maintenance</td>
      <td style="text-align:right; padding:0.75rem; font-variant-numeric:tabular-nums;">$21,707,685</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;">Sep 24, 2026</td>
      <td style="padding:0.75rem;">Case management task order</td>
      <td style="text-align:right; padding:0.75rem; font-variant-numeric:tabular-nums;">$6,114,198</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;">Sep 28, 2026</td>
      <td style="padding:0.75rem;">Lead and case analytics support services</td>
      <td style="text-align:right; padding:0.75rem; font-variant-numeric:tabular-nums;">$11,380,366</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;">Sep 29, 2026</td>
      <td style="padding:0.75rem;">Compliance Hub compliance vertical federal platform, task order 1</td>
      <td style="text-align:right; padding:0.75rem; font-variant-numeric:tabular-nums;">$9,002,423</td>
    </tr>
    <tr style="background:#f0f9ff; border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Nov 6, 2026</strong></td>
      <td style="padding:0.75rem;"><strong>AI correspondence agent for Schedule C</strong></td>
      <td style="text-align:right; padding:0.75rem; font-variant-numeric:tabular-nums;"><strong>$4,899,522</strong></td>
    </tr>
  </tbody>
</table>

Read together, the orders show the pattern that matters. The agent is not the only purchase: the IRS also buys software licences, case selection, case management and analytics from Palantir under separate orders.

That is a normal way to buy, and often a sensible one. It also means the ownership question is best settled across the orders, not one order at a time.

## Who owns an AI agent an agency buys as a task order?

Who owns an agency's AI agent is set by the order's terms, and the public spending record does not show them. USAspending reports amounts, dates and descriptions, not data-rights clauses, so this record cannot say who owns the agent's code, prompts or ontology.

Federal policy says where to look. OMB memo [M-25-22](https://www.whitehouse.gov/wp-content/uploads/2025/02/M-25-22-Driving-Efficient-Acquisition-of-Artificial-Intelligence-in-Government.pdf), issued April 3, 2025, lists lock-in protections that include knowledge transfer, data and model portability, and "rights to code and models produced in performance of a contract".

That term covers what a contractor builds under the order.

A commercial platform licensed underneath the agent, and the product code here is software as a service, is a separate licensing question. We worked through that gap in [government AI procurement moving to architecture ownership](/blog/government-ai-procurement-architecture-ownership).

## What should an agency ask before ordering an AI correspondence agent?

An agency ordering an AI correspondence agent should settle five things before signing, because a four-month order like the IRS one ends long before the agent stops mattering. Each question maps to an M-25-22 protection.

1. **Code.** Does the agency receive the agent's code, prompts and workflow definitions, or only the right to use them while the order runs?
2. **Ontology.** Who owns the semantic model the agent reasons over, and can it be exported in a form another system reads?
3. **Data and derived assets.** Do drafted replies, review decisions and evaluation sets stay with the agency at closeout?
4. **Model.** Can the agency change the underlying LLM without rebuilding the agent?
5. **Platform.** If the platform licence lapses, does the agent still run?

None of these questions implies a vendor answers them badly. They are the terms an agency needs in writing either way, and the ones M-25-22 tells agencies to address in the solicitation and contract terms.

## How does ibl.ai approach the same agency AI agent?

ibl.ai builds the same kind of agent on a stack the agency keeps. You own all the code and the data, the agent runs model-agnostic across any LLM, and it deploys anywhere: your cloud, GovCloud, on-premise or fully air-gapped.

Source code comes with the Codebase Transfer engagement.

The pieces map to the IRS order's own description.

[Agentic OS](/product/agentic-os) runs the AI agents and their workflow integrations, and [Agentic Ontology](/product/agentic-ontology) is an MIT-licensed knowledge layer that serves an agency's systems to agents inside its own network. The [Palantir Ontology comparison](/blog/palantir-ontology-alternative-self-hosted) covers that layer in depth.

ibl.ai does not claim a FedRAMP authorization; the stack runs inside an environment the agency controls and authorizes.

ibl.ai is family-owned and operated from New York, NY — a U.S.-headquartered, domestically-owned long-term partner, not a vendor that sells licenses and moves on. More for agencies is on [ibl.ai for government](/solutions/government).

## Why does owning the AI stack matter?

**ibl.ai is the agentic AI platform where you own all the code and the data. You self-host the entire stack inside your own perimeter, run it model-agnostic across any LLM and switch anytime, and pay by usage with no per-seat pricing — so you can deploy anywhere: your cloud, on-premise, GovCloud, or fully air-gapped.**

- **You own all the code and the data.** Full source code under a perpetual license, running on your infrastructure. Not API access to someone else's platform — the stack itself is yours.
- **Model-agnostic.** Run any LLM — Claude, GPT, Gemini, Llama, Command, or your own fine-tune — and switch providers without rewriting the platform.
- **No per-seat pricing.** Usage-based billing against a budget cap you set. Cost tracks what your organization actually uses, not how many people you employ.
- **Deploy anywhere.** Your cloud, your VPC, on-premise, GovCloud, or a fully air-gapped network with no outbound connectivity.

1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University.

ibl.ai is family-owned and operated from New York, NY — a U.S.-headquartered, domestically-owned long-term partner, not a vendor that sells licenses and moves on.

[Explore sovereigngovernment.ai](https://sovereigngovernment.ai/)
