In 1983, perhaps the dangers of the Cuban Missile Crisis resurfaced when two similar events occurred in the same year, but this time not because of deterrence and Cold War politics, but due to a major data-driven software warning system crisis.
The Soviet Union, in its endeavour to bolster its nuclear deterrence and warfare capabilities, launched Operation RYaN (Raketno-Yadernoye Napadenie), a large, automated early-warning system used by the Soviet KGB. It relied on algorithms to assess the likelihood of a U.S. first strike by analysing 40,000 political, military, and economic indicators.
The Nuclear Crises
The real crises began to unfold when NATO launched its own nuclear exercise, known as Able Archer. During the realistic Able Archer 83 command post exercise, the RYaN algorithm treated the simulated changes as real-world threats. Because the computer lacked contextual awareness, its output triggered systemic paranoia, forcing the Soviet military to ready its nuclear arsenal for a preemptive strike. It was prevented because human intelligence officers, like USAF Lieutenant General Leonard Perroots, ignored rigid escalation protocols and trusted human intuition over algorithmic warnings.
The Danger
Recently, a report revealed that due to AI’s misjudgment and overreliance on AI-enabled military software, combined with stale, outdated intelligence, it led to a catastrophic precision missile strike on the Shajarah Tayyebeh Elementary School in Minab, Iran. It resulted in the deaths of over 150 civilians, including 123 children. According to reports, multiple factors led to flawed judgment on the AI’s part, including an outdated database that identified the school as an IRCG location because it was built on part of the IRCG naval compound. But the main flaw was overreliance on speed, which is too fast for independent human verification. Even worse is the bias and blind trust human operators place in the AI, especially in crises or situations of intense pressure, as seen during the Archer crises in 1983 and similarly in Minab strikes, where AI data and target marking didn’t undergo independent human verification and were left to the AI. Human operators mistakenly expected the software to flag stale data or inconsistencies, which AI may or may not do.
Crises For Spies
The real danger lies in security and intelligence establishments, where decision-making spans are gradually shrinking amid the evolving threat landscape and rising clandestine activity. Intelligence agencies and security establishments are also caught in “flash wars”, where technology is further complicating and compressing decision-making time, creating a sense of automation and reliance on AI technologies such as the Pentagon’s use of Palantir’s Maven Smart System (which incorporates advanced AI tools, including Anthropic’s Claude) to process information and drastically compress target selection timelines. The shift from HUMINT to TECHINT and ELINT is rapidly growing, but at what cost? The consequences are devastating, as seen in Minab or in putting nations on the brink of doom in 1983. The obsession with tech and automation is not new; it has been running for decades, with the CIA running secret projects and whatnot to achieve information superiority, but that obsession did come with great strategic costs or Intelligence failures.
CIA’s Tech Obsession
CIA’s failure in Iran’s 1979 revolution that overthrew Shah’s government was a result of overreliance on technical intelligence and less on Human Intelligence. In the 1960s and 70s, the U.S. heavily invested in electronic listening posts in northern Iran to spy on the Soviet Union. Content with the technical data flowing through these stations, the CIA grew complacent regarding internal Iranian politics.
The fatal bombing of the Chinese Embassy during the 1999 NATO bombings in Yugoslavia occurred when a U.S. B-2 bomber dropped five laser-guided bombs on the Embassy of the People’s Republic of China in Belgrade, killing three Chinese journalists and injuring 20 others. The strike triggered a massive international diplomatic crisis. It was planned using advanced geospatial targeting methods, relying heavily on satellite imagery and computerized military grid coordinates to locate the Yugoslav Federal Directorate for Supply and Procurement (FDSP). Even in this case, Technical Intelligence and assessments were given more weight than Human Intelligence. U.S. intelligence agencies suffered from a severe systemic bias: they assumed their digital targeting databases and computer-generated maps were absolute truth, and no independent on-the-ground verification by Human Intelligence was ever conducted before carrying out targeted strikes in a conflict-ridden state.
According to reports, the CIA’s electronic database was using an uncorrected, outdated map. The FDSP building had never been there, and the Chinese Embassy had moved to the designated grid coordinates years prior. Apart from false marking or targeting, the Intelligence agency exposed itself to deception vulnerability due to blind trust in TECHNIT. The prime example is the Vietnam War’s Tet Offensive in 1968. The U.S. heavily relied on the “McNamara Line”, a multi-million dollar network of acoustic and seismic sensors dropped from aircraft to electronically monitor troop movements along the Ho Chi Minh Trail. Leading up to the offensive, electronic sensors picked up massive troop movements and heavy artillery placement concentrated around the isolated U.S. Marine base at Khe Sanh. U.S. commanders, looking strictly at electronic telemetry and data sheets, became convinced that Khe Sanh was General Giap’s ultimate objective.
Because the U.S. command trusted the electronic sensor spikes blindly and lacked reliable human networks inside the Viet Cong command structure, they shifted critical reserves to defend the base. This left the major cities completely vulnerable to the real surprise onslaught, shattering the official narrative that the communist forces were weak and incapable of major operations.
The Real Problem
The urge for pre-emptive action and the threat of an overwhelming reaction from the adversary’s side, sometimes in zero time, pushes a rush in decision-making. Beyond this flash-war excuse, four primary factors drive this AI problem in Intelligence. First is automation bias: the belief that AI-processed technical Intelligence inputs are mostly true and accurate. Second is ignoring Technical Intelligence hallucinations; AI sometimes hallucinates and generates off-track Intelligence that directly contradicts situational realities, showcasing a failure to understand the intent. Third is the susceptibility of Technical Intelligence, or AI, to deception; it is highly vulnerable to spoofing, camouflage, and radio silence. Fourth is the algorithm trap, which gives a limited reading of the situation. Intelligence agencies and the security establishment are trapped by algorithmic parameters and struggle with cultural nuances, irony, or code words, leading to a complete misreading of the situation, as the CIA failed to anticipate the 1979 Iran revolution. The danger escalates further when human intelligence, already blinded by bias, is compounded by technical or artificial intelligence, leading to more conflicts and fatal events.
The Hybrid Architecture
HUMINT may deliver street reality, but it takes time, and the coming age of flash wars will gradually reduce that time. However, HUMINT will still be strongly relevant because it is the only way to understand your adversary’s real intent, identify the complete picture, and detect possibilities of deception, in which AI may struggle. Without HUMINT, AI operates in a context vacuum and often falls victim to automation bias. The debate between speed and accuracy can be resolved only through a hybrid mix of HUMINT and TECHNIT, where HUMINT provides accuracy and in-depth detail, and TECHNIT supplies data and analyses actions. To maintain high operational tempo, military and intelligence agencies must blend these disciplines without letting AI speed outrun HUMINT’s verification capabilities. The hybrid mix can be prepared by integrating HUMINT into AI Intelligence processing, basically HUMINT-driven AI, which runs on HUMINT inputs, assessments, and periodic data and Intelligence updates in AI by Intelligence officials. The Natural Language Processing (NLP) models constantly parse decades of handwritten agent reports, cultural atmospheric briefs, and diplomatic cables.
Throughout the process, there has to be a Human verification system, apart from improving accuracy through data and intelligence feeding. The AI judgements/reports must be reviewed through a thorough Intelligence cycle, involving filtering and analysis. A circuit-breaker kind of setup can be utilised in this case. If a system identifies a target that deviates from historic patterns, or if it crosses into highly populated zones, the system must automatically lock and require a review of AI-assessed intelligence. An architecture of HUMINT and TECHNIT in the Intelligence cycle is a need of the hour, as flash wars in the coming time will be more devastating and fatalistic if dependency on AI increases and HUMINT decreases, because even a margin of error can cause hell to break loose.
