Thu 2 April 2026

Thomas Mulligan, a former CIA case officer and RAND researcher, argues persuasively that the proliferation of generative AI, particularly large language models, will paradoxically elevate the role of human intelligence (HUMINT) in espionage. As AI democratizes disinformation—enabling agents to fabricate convincing backstories or false intelligence—it erodes trust in digital channels, pushing operatives toward analog methods like dead drops and face-to-face encounters. While AI offers tools for pattern recognition in surveillance or psychological manipulation, Mulligan warns it also heightens risks for human handlers, making counterintelligence more vital. This perspective revives Cold War-era spycraft, but skeptics might question whether it underplays AI's potential to automate even interpersonal deceptions, potentially rendering human elements as vulnerable relics in an increasingly synthetic intelligence landscape. Overall, Mulligan's analysis underscores a timely shift: in an era of deepfakes, the human touch may indeed be the ultimate verifier of truth. Key points: • AI enables easier fabrication of false intelligence, increasing the need for human verification in HUMINT. • Digital communications lose trustworthiness due to AI-generated disinformation, favoring old-school spycraft like dead drops. • AI tools can aid spies in manipulation and data analysis but also pose detection risks for operatives. • Human intelligence officers will remain essential as AI complicates distinguishing fact from fiction. Expert question (counterfactual): What if rapid advancements in AI-driven behavioral prediction and real-time deepfake countermeasures ultimately automate HUMINT tasks, making human spies as obsolete as they were feared to be?

Sovereignty is the key question in AI. Can we realistically have sovereignty by country or region in a world where infrastructure investment of US BigTech runs to $500B / year. The scale of compute, data, talent and distribution seem to be key, and that is not something that is easily addressed.

ServiceNow's push into agentic AI frames autonomous workflows as the next frontier beyond copilots, promising to redefine enterprise work through self-acting software, but only with rigorous governance to manage risks like data access and outcome drift. Executives advocate treating AI agents as a distinct identity class—neither fully human nor machine—equipped with least-privilege permissions, starting implementations in controlled IT domains like email triage to demonstrate tangible ROI, such as slashing resolution times by 13%. Tools like the AI Control Tower enable runtime monitoring and traceability, ostensibly safeguarding PII and ensuring alignment with intent. While this narrative aligns with ServiceNow's platform-centric strengths and addresses real security fault lines, it warrants skepticism: the hype around productivity gains may overlook integration complexities and the historical struggle of enterprises to monetize AI, potentially positioning governance as a compliance burden rather than a velocity booster. Key points: • Treat AI agents as a new identity class with distinct, scoped permissions for data access. • Begin agentic AI deployments in narrow IT use cases to measure ROI, like reducing task times from days to minutes. • Employ AI Control Tower for logging actions, tracing outcomes, and monitoring drift to maintain security and traceability. • Prioritize governance and security to enable scaling, as lapses could stall AI programs amid CIO pressures for value realization. Expert question (counterfactual): What if robust governance, while mitigating risks, inadvertently creates bureaucratic hurdles that hinder the rapid experimentation needed for agentic AI to truly outpace human workflows in diverse enterprise domains?

Benjamin Laker's column astutely delineates the boundaries of AI in managerial practice, advocating its use for accelerating rote tasks like data synthesis and drafting while cautioning against outsourcing human-centric judgments on values, relationships, and trust. This balanced perspective counters the hype surrounding AI as a panacea for productivity, emphasizing that overreliance risks eroding critical thinking and emotional intelligence—core to effective leadership. However, Laker's framework assumes managers possess the self-awareness to discern these boundaries consistently, a skepticism warranted given evidence of cognitive biases in tech adoption. By positioning AI as a tool rather than a proxy, the piece promotes deliberate integration, potentially fostering more resilient organizational cultures amid rapid technological shifts, though it underplays the need for systemic training to prevent unintended delegation of authority. Key points: • Use AI for mechanical tasks like summarizing reports or outlining agendas to free up time for strategic sensemaking. • Avoid AI in decisions involving trust, values, or relationships, such as delivering feedback or assessing candidate resilience. • Treat AI outputs as raw material requiring human editing to maintain judgment and accountability. • Instruct AI to generate counterarguments to challenge biases and broaden decision-making perspectives. Expert question (counterfactual): What if advancing AI capabilities in emotional intelligence and contextual nuance render the distinction between mechanical and human tasks obsolete, forcing managers to redefine leadership entirely?