What’s actually in the new EU AI Act amendments?

Claims the sources don't agree on — every side, shown side by side.

The headline number is real and the strategic shift it represents is meaningful — a $2.1B autonomous-systems award is more than three times the next-largest comparable contract in Anduril’s history, and the swarm-coordination capability is the first of its kind to reach production scope at the DoD.

But the deployment math matters more than the dollar figure. Even with an aggressive timeline, INDOPACOM fielding in Q3 2027 represents 18 months of integration work, doctrine writing, and operator training before any systems become combat-relevant. Historical precedent for AI defense contracts suggests another 12–24 months on top of that before measurable operational capability.

The protest filings from RTX, Lockheed, and Boeing are a separate signal. Defense primes don’t file protests on contracts they expect to lose cleanly — they file when the source-selection paperwork is thin enough that a GAO inquiry could either reverse the award or carve out subcontract concessions. Either way, the headline scope rarely survives a protest unchanged.

What’s actually new here is the validation of autonomous swarms as a procurement category. Anduril’s contract is less interesting as a single transaction than as a forcing function: other primes will now have to ship comparable capability or cede the category permanently.