The Pentagon's Pacific command embraces AI for fear of falling behind, and its biggest risk already produced a near-miss in September

By Honolulu Star-Advertiser · November 7, 2026.

U.S. Indo-Pacific Command (PACOM) is rolling out artificial intelligence across the entire command, its AI chief, Scott Brunstetter, said at the TechNet Hawaii conference on October 28. Brunstetter said Adm. Samuel Paparo ordered its adoption throughout the structure. Brunstetter's title is now "Chief Super Intelligence Officer," because an executive order from Donald Trump requires government agencies to call AI "superintelligence" (SI).

The official justification is one of relative risk. According to Brunstetter, the command decided that "the operational risk of not adopting AI exceeds the fundamental risk of adopting it." Uses range from logistics to intelligence analysis. Tagup, a company that works with U.S. Marine Corps Forces, Pacific, says it can already iterate plans faster than a staff officer. It also uses AI to track the expiration dates of medical supplies.

On surveillance, Rear Adm. Jeffrey Buschmann said PACOM tracks at least a thousand moving targets every day. According to him, it almost always has "100% control" over those belonging to Russia and China. The figure comes from the command itself and has not been independently verified. His stated concern is what comes next: prioritizing targets and choosing the most efficient weapon in the event of a conflict.

The counterweight came in September. According to CNN, elite troops in the Indian Ocean were about to board a Chinese cargo ship based on an intelligence report saying it was carrying nuclear components. Just before the operation, an analyst at Special Operations Command discovered that the report had been produced with the help of AI and misidentified the cargo. A source cited by CNN called it "completely false" and said it "almost started a war." This is the account of an anonymous source, not a fact we have verified.

Another problem is physical. The models live in data centers, and the command does not want its units in Hawaii to depend on those on the mainland. Ricky Buchanan, PACOM's chief of intelligence information systems, put it this way: they cannot rely on a tool that depends on continental data centers. Meanwhile, the military is building a 56,000-square-foot (about 5,200 m²) lab in Kihei, facing opposition from residents, environmentalists and Native Hawaiian groups. Buchanan is also skeptical of mobile data centers: he does not want to deploy "a juicy target." The article notes that data centers have already been attacked in active wars, such as Amazon's in Bahrain and the UAE and several in Russia.

**Our take.** The command's thesis, "not adopting is riskier," is reasonable and almost inevitable in a race with China. But it is an argument that can justify any deployment just as well. What matters is the controls that come with it, and the cargo ship case shows that human oversight worked by sheer luck: one analyst who checked in time. That is not a system; it is a gamble.

Brunstetter says they do not blindly trust the tools and that analysts must understand how they work. That is the right attitude, but there is a tension: AI is sold on speed, and human review takes time. As the pace of decisions increases, so will the pressure to skip that verification. Broadly, this is the pattern of every rapid agentic adoption: deployment runs ahead of auditing and traceability.

It is also notable that AI is becoming critical infrastructure with a material dependency: electricity, water and data centers that are, at the same time, military targets. That strengthens the case for efficient models that run locally, an area where the open frontier is advancing quickly.

In the short term, the militarization of AI brings real risks of error and escalation. In the long term, the same analytical and logistics capabilities can be applied to healthcare, supply chains and crisis management. The underlying challenge is not to halt the technology, but to require those who deploy it to show how they verify its outputs before a failure costs lives.