Yesterday, the Washington Post reported that the United States and Russia acted to strip oversight from a global AI weapons pact.

According to the Washington Post article, the United States and Russia used lawyers and at the United Nations to remove various safeguards around the use of artificial intelligence in weapons. Some of the language the two countries wanted removed involved requiring the systems to operate in a predictable and reliable manner and removed a clause mandating ethical considerations be taken into account when using AI weapon systems (Verma 2026).

The Post article cited Verity Coyle, who said, “It cold mean machines can make life-and-death decisions without human control. The practical impact is stark: more civilian harm, less accountability and a faster slide into riskier, automated warfare” (Verma 2026).

The story notes that these efforts to limit the capabilities of AI systems come amid a growing wave of incidents where AI bots have autonomously hacked the servers of other organization, sparking fears the technology could become uncontrollable (Verma 2026).

DEAD COGNITIVE LABOR AND VERIFICATION LAG

Some of the issues this brings forward are dead cognitive labor, as well as verification lag.

Dead cognitive labor is collective intelligence, detached from the living social relations that created it, stored inside computers, that can then be reactivated at a later date … sometimes at the push of a button. This storage is not just data and fact, it is stored judgement. Because dead cognitive labor is the stored knowledge that is learned from trial and error, acquired not from just a few mishaps, but by scraping the mistakes of thousands (or millions) of stored judgments over decades (Van Pelt 2026). And the reactivation of this stored judgement happens at machine speed. The problem this creates is two-fold.

First, the sheer speed these judgements can occur. What may take hours or weeks of debate before deciding to make a military action can happen within a few seconds. But that data is only as good as the accuracy of the data. Making decisions in seconds on perfect data might be okay. But if the data is not accurate, then the result could be bombing a school.

Second is the issue of verification lag. Mistakes happen, but the power of verification lag is that the decision can be acted upon, but then the victims of that decision may need weeks or months to contest the decision (Van Pelt 2026). Whether it is striking the wrong target, or accidentally denying someone medical benefits, those decisions are felt by living human people. AI operates with judgements created by humans, without the same risk of punishment if they get it wrong.

Without live human oversight of AI weapons systems, verification lag can be a real problem for innocent people. The wave of AI hacking incidents is one example of this threat … without the physical problem of bullets.

References:

Van Pelt, Craig. 2026. “Dead Cognitive Labor and the Tempo of Machine-Time: AI, Infrastructural Capture, and the Privatization of the General Intellect.” Sociation, 26(2), 19-34.

Verma, Pranshu. 2026. “How the U.S. and Russia Weakened a Global Effort to Regulate Killer AI.” Washingtonpost.Com, The Washington Post. Retrieved September 27, 2026 (https://www.washingtonpost.com/technology/2026/09/26/how-us-russia-weakened-global-effort-regulate-killer-ai/).