Recent news about AI inside the Pentagon shows a concern I raised in Small Wars Journal on March 30, in “Corporate AI as the Military’s Weakest Link.” The concern was ordinary people becoming too comfortable with commercial AI inside military workflows. Contractor-managed environments, excessive permissions, and informal data sharing were already reasons to slow down and examine what was being built. Putting a convenient tool into daily use changes how people behave once it becomes part of standard operating procedure. The paradox of accountability is expecting security discipline to hold while everyone gets comfortable in the warm embrace of the corporate AI industry.

According to WIRED’s October 29 report, an individual allegedly hooked their personal Claude Code account into a Pentagon system used by multiple intelligence agencies and holding sensitive, unclassified information. The agent reportedly ran for two months before the problem was identified. WIRED’s sources say it apparently sent credentials and a network map to an outside site NIST had already flagged as high risk. The individual’s employer disputes the allegations and says an independent investigation found them without merit and that it met its obligations. No classified-network compromise has been established, nor does this article say it does. This is about concern and governance, not any individual company or incident.