After AMAC: The New Bar for AI in Federal Acquisition

On April 13, AFRL cancelled FA8652-26-R-0001, the $10 billion AMAC IDIQ, six weeks after proposals were due. The one-page notice stated the Government needed to reassess its requirements and acquisition strategy. The cancellation followed the dismissal of three GAO protests on April 6.

No award. No evaluation. Dozens of offerors, primes and small businesses alike, now hold proposal content they cannot use. This is a $10 billion IDIQ. The defense industrial base did not spend millions responding to it. It spent tens of millions, almost certainly hundreds of millions, across bid and proposal budgets, color team reviews, past-performance write-ups, pricing builds, teaming negotiations, and legal review. That is capital that should have been funding new technologies, new weapons systems, and the research the warfighter will need in the next decade. Instead it was burned on an acquisition that will not exist.

Public commentary has zeroed in on the surface-level drivers of the reassessment. Award criteria weighted heavily toward entrenched incumbents. Prime-level experience requirements applied even to small-business offerors. Subcontracting commitments that were difficult to meet in combination with the experience bar.

Those are real. They are also symptoms, not the disease.

The disease is simpler, and more uncomfortable. The AMAC solicitation was not authored the way serious federal solicitations have been authored for sixty years, by a contracting workforce that owns the outcome, knows what a protest looks like, and is trained to write requirements that will hold. It was generated by a thin LLM wrapper by individuals masquerading as AI experts while standing up a chatbot on top of a general-purpose language model and trusting it to help draft a $10 billion procurement.

That is the root cause. The lopsided experience criteria. The incoherent subcontracting math. The evaluation logic that could not survive first contact with a protest. Those are not three independent acquisition-design mistakes. They are the characteristic output of an ungoverned language model generating confidently inside a workflow no one built to catch its mistakes. No source tracking. No policy alignment. No audit trail a contracting officer could reconstruct. A $10 billion solicitation authored by autocomplete, pushed to SAM.gov, and treated as a real acquisition until it collapsed under the weight of what it actually was.

When a solicitation is drafted that way, the cost is not abstract. It is paid by every offeror who responds in good faith. In AMAC, it was paid twice. First in the tens of millions the defense industrial base spent building responses against requirements that would not hold. Then again when the acquisition was cancelled, that work evaporated, and the R&D those dollars should have funded, the next drone, the next counter-UAS system, the next autonomy stack, went unbuilt.

That is the lesson underneath AMAC. That is what makes this moment a dividing line. And that is what the GovCon community will not forget the next time an AI-generated acquisition artifact hits the street.

Speed without defensibility is not transformation

As AI moves deeper into federal acquisition, the tempting narrative is that speed is the transformation. Faster intake. Faster evaluation. Faster time-to-award.

Speed is only half the answer. Defensibility is the rest.

Every acquisition decision leaves a paper trail. That paper trail has to hold up to protest. It has to hold up to an Inspector General review. It has to hold up to GAO precedent and to the plain-language test of whether the solicitation does what it claims to do.

AI that moves procurement faster but weakens the paper trail is not a win. Unsourced market research, untracked document revisions, and artifacts no one can reconstruct six months later are not productivity gains. They are liability compounding at machine speed. They are prose, not provenance. They are the same failure mode that cost the defense industrial base a nine-figure bill on AMAC.

AI built for defensibility, with version tracking, citation-backed market research, traceable artifact generation, and GAO-precedent awareness, compounds in the other direction. Every decision becomes easier to defend. Every artifact becomes easier to audit. Every downstream protest becomes easier to answer.

That is the bar. AMAC just raised it.

What transparent, auditable AI looks like

At Rohirrim, we built the Unified Acquisition Platform™ around three non-negotiables: security, speed, and precision. An acquisition artifact is only as useful as it is defensible.

We build the guardrails first, then the race car. That ordering is not optional in federal acquisition, and AMAC is the most public reminder of why.

That shows up in how UnifiedAcquire™ works. Every document carries a complete audit trail. Market research records cite back to source. Statements of Work, Acquisition Plans, and Sole Source Justifications generate inside FAR and DFARS-aligned templates the contracting workforce already recognizes. Role-based access control governs who can see what and when. The knowledge base is single-tenant and purpose-built for the organization that owns it. Nothing a contracting officer approves inside UnifiedAcquire traces back to a black-box generation event. Every word has a source line. Every artifact has a hash. Every decision has a path back to the record.

That is what defensible AI in acquisition actually looks like. The difference between UnifiedAcquire and a language model wrapped in a user interface is the three years of domain-specific data engineering that sits underneath it. One compounds trust. The other compounds risk, and AMAC is what that risk looks like when it lands.

The wave that follows AMAC

When AFRL re-issues AMAC, and when adjacent AFRL, AFLCMC, and Space Force vehicles reach industry day, the standard for AI-enabled acquisition tooling will be higher. Program offices watching this episode will ask harder questions. What is the audit trail? Who owns the data? Where does market research come from? What guardrails govern the model? How does the tool handle the experience-criteria and small-business logic that gets scrutinized in protest? Can the platform justify every step to a GAO attorney?

These are the right questions. They should have been the first questions.

AI in acquisition cannot just be faster. It has to be defensible. It has to be transparent. It has to be governed. And it has to be built for the acquisition workforce that answers for every decision, through protest, audit, and beyond.

The category of tool that produced the AMAC solicitation is the category the market is about to move past.

Speed of Need. Built to withstand scrutiny.

AMAC did not change what we build at Rohirrim or what UnifiedAcquire™ delivers today. It defined what the market will expect moving forward.

The Department of War Acquisition Transformation Strategy set the direction: acquisition must operate at the Speed of Need. Our platform was built for it. But speed only counts when the work holds. An award that collapses six weeks after proposals were due is not faster acquisition. It is a wasted acquisition process, a nine-figure bill paid by the defense industrial base, and R&D that never happened because the capital was tied up responding to a document an ungoverned “chatbot app” was allowed to write.

The organizations leading the next decade of federal procurement will treat transparency and speed as one standard, not a trade-off. Rohirrim is built for that standard.

Steven Aberle

CEO