CENTCOM Patches the Maven Smart System Instead of Retiring It After Minab [Update]
US Central Command has rewritten how artificial intelligence feeds its lethal targeting decisions, adding civilian-movement data feeds, rebuilding the workflows that refresh and vet planned targets, and pushing dozens of upgrades into the Maven Smart System built by Palantir Technologies. The adjustments had not been disclosed before. They respond directly to the February strike that destroyed the Shajarah Tayyebeh Elementary School in Minab, Iran, killing at least 157 people, 123 of them children. CENTCOM has not retired its AI combat targeting stack. It has patched the data going in and the human checks around it.
The changes came days after the Pentagon review we covered earlier this week, which traced the planning failure to overreliance on Maven's output, stale intelligence and deep cuts to the teams responsible for limiting civilian harm. The review followed the February 28 strike, which came during the opening wave of the campaign against Iran. CENTCOM's response attacks the first two problems with engineering and process. The third it leaves largely untouched.
What CENTCOM Actually Changed
Three workstreams sit at the center of the overhaul, and each maps to a gap the strike exposed.
- Target vetting and refresh: CENTCOM rewrote the processes used to keep planned targets current and to verify them before release. Palantir has upgraded Maven so the system re-examines the intelligence underneath a target, identifies factors that should disqualify it, and flags inconsistencies and inaccuracies a human reviewer may have missed.
- Civilian-movement data: New open-source feeds now track how civilians move in and around potential targets. The Minab school stood across from a site that had been nominated for attack, a proximity the earlier pipeline did not resolve.
- Maven upgrades: Dozens of changes have been pushed into the battle-management platform itself, touching how it surfaces, scores and re-checks candidate targets.
Each element addresses a different link in one chain. The vetting rewrite handles stale intelligence. The data feeds handle population awareness. The Maven upgrades handle the machine's own error detection. Taken together they are an admission that the combined human and machine pipeline behind the Minab strike package was inadequate, without conceding that the software should leave the loop.
CENTCOM has acknowledged publicly that it used advanced AI tools during the Iran campaign, while stopping short of attributing the Minab casualties to any single system. The disclosed changes are narrower than a withdrawal and broader than a patch. They alter what data Maven sees, how it re-examines that data, and which humans sign off before a target is released.
The framing matters for what comes next. The review spread responsibility across tooling, stale data and thin staffing rather than naming a single point of failure. That keeps the platform in the inventory and shifts the argument to procurement terms. The question defense buyers now inherit is what a targeting system must prove about its own outputs before planners are permitted to lean on them.
| Failure identified in the review | Fix now in place | What remains open |
|---|---|---|
| Stale intelligence on planned targets | Rewritten target-refresh and vetting workflows | Verification still depends on analyst capacity |
| No visibility of civilians near nominated sites | New open-source civilian-movement feeds | Open-source feeds lag real-time occupancy |
| Human review missed disqualifying factors | Dozens of Maven upgrades; automated re-review of underlying intelligence | Alerts still need people to act on them |
| Civilian harm mitigation staffing cut by about 90% | Not addressed by the disclosed changes | Fewer than 20 personnel remain |
Fixing the Maven Smart System With Itself
The most consequential detail is structural. CENTCOM is asking the system that contributed to the error to catch its own errors. Maven compressed target preparation that once consumed hours into minutes, and that speed is why the platform exists. During the opening 24 hours of the Iran campaign, the software generated hundreds of strike coordinates and US forces hit more than 1,000 targets. Every added review loop runs against the tempo that made the tool worth buying.
That tension defines the trade-off CENTCOM has chosen. A slower, more heavily documented targeting cycle lowers the chance of another Minab. It also narrows the operational advantage Maven was procured to deliver, under a Pentagon contract valued at up to $1.3 billion. The command is betting that automated re-review absorbs most of the extra caution and leaves human analysts to adjudicate only the flagged cases.
Palantir's exposure runs through the same agreement. The upgrades expand what the platform is expected to do rather than shrink its role, which hands the vendor a compliance story for future bids and a new obligation to show that the internal check holds up under load. If the flagging works, re-verification becomes a differentiator against rival battlefield AI products. If it does not, the failure is attributable to the software in a way the original review deliberately avoided.
The bet is testable. If the upgraded flagging inside the Maven Smart System reliably surfaces disqualifying factors, the accountability layer becomes a selling point for Palantir as it markets the platform to other defense customers. If flags arrive faster than analysts can process them, the upgrades add latency without adding judgment.
The Limits of Automated Review
The review's central finding was an expectation mismatch. CENTCOM personnel assumed Maven would flag stale or inconsistent intelligence on its own, and as configured it did not. That gap between assumed and actual capability is what the upgrade cycle now targets, and it is harder to close than a data-feed problem.
Automated skepticism has to be designed in. A system tuned to move planners quickly through hundreds of candidate targets is, by construction, not tuned to slow them down. Adding disqualification checks and inconsistency flags changes the product's purpose at the margin, which is why the change had to be written into both the software and the human workflow rather than switched on inside the Maven Smart System alone.
The review also found that civilian harm mitigation teams had been reduced by roughly 90%, leaving fewer than 20 personnel across several units. Automation does not restore that capacity. An alert nobody has time to investigate is functionally the same as no alert, and the reduced headcount that contributed to the original failure is still in place.
For defense planners, the operative question is whether automated re-review substitutes for staffing or supplements it. CENTCOM's disclosed changes treat the software layer as the faster fix. That ordering is defensible in the short term and exposed over a long campaign, because target volume, not algorithm quality, is what overwhelms review capacity, and volume is what Maven is designed to raise.
Other militaries watching the case will draw the same inference. An AI targeting stack that ships with built-in re-verification is easier to defend in public and in court than one that leans on thin human checks. The re-verification features now in place are likely to shape procurement requirements for battlefield AI beyond this contract, while the staffing arithmetic stays a budget question rather than a product one.
The next test is a documented one. CENTCOM has not disclosed how many targets the upgraded flagging has blocked, or how often analysts overrode a flag and released a target anyway. Those two figures, rather than the count of upgrades, will show whether the targeting pipeline has actually changed.
Why this matters
For any organization that deploys AI to generate high-volume recommendations, Minab is the clearest case yet of what happens when machine output outruns the human capacity to check it. CENTCOM's answer was to make the machine re-check itself rather than to add people, and that choice will be copied far beyond defense, wherever fast AI output meets thin review teams. The measure of the overhaul is whether enough people remain to act on the flags Maven raises.
Related Articles
- Palantir's Maven Program of Record Status Locks In $2.3B in Pentagon Funding
- Palantir Ammunition Tracking Contract: Army's $48M Bet on One Platform
- Pentagon approves AI military targeting doctrine shift
✔Human Verified
Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.