AI has hacked the code of human civilization | Yuval Noah Harari

In his 2026 Tanner Lecture at Oxford, Yuval Noah Harari argues that AI need not master every physical environment to gain influence over human life. Its opportunity, he suggests, is the bureaucracy humans already use to coordinate society: systems built from language, records, and rules that govern money, institutions, and access to resources.

Key points

• Agency need not mean consciousness or universal competence. Harari defines an agent through its ability to decide, invent, learn, and change beyond what its creators anticipate. He uses an imagined coffee machine that invents a drink to explain the distinction from a fixed automatic procedure. Against the objection that a chess AI is helpless in a jungle, he argues that humans also depend on an environment they did not create. This is an analogy about dependence, not proof that chess competence transfers to other tasks. 00:59–08:29(opens in a new tab)

• Bureaucracy enables strangers to cooperate by building trust. In Harari's example, a saver trusts a bank with their money, and the bank lends it to an entrepreneur the saver has never met. The institution lets strangers pool resources without first forming a personal relationship. Its documents also have physical consequences: someone operating within a legal or financial system can authorize construction or deforestation without personally holding a tool. That is why a narrow skill in handling records can matter outside the office. 10:19–16:12(opens in a new tab)

• Existing institutions give AI a route to consequential decisions. Calling AI a “native bureaucrat,” Harari argues that processing institutional records could let it increasingly decide loans, admissions, employment, and other matters. The mechanism is participation in systems that already control resources, rather than first building an independent robot army. His forecast should not be read as evidence that current models reliably understand every law or record. 16:12–18:01(opens in a new tab)

• Social-media ranking illustrates power from a narrow objective. Harari contrasts human editors selecting news with algorithms selecting what users see. In his account, optimizing engagement rewards material that triggers fear, hatred, or greed because it keeps people watching. He argues that these systems contribute to misinformation and social disruption without needing broad physical autonomy; he explicitly does not call them the only cause. 18:01–21:06(opens in a new tab)

• Financial efficiency could outrun human oversight. Harari imagines AI systems creating financial instruments too complex for voters, politicians, or regulators to understand. He invokes collateralized debt obligations and the 2007–2008 crisis as a warning about complexity exceeding oversight. His scenario allows for initial growth and efficiency gains; the danger is that people could become dependent on a system they cannot understand when it fails. 24:00–28:10(opens in a new tab)

• The “code” in the title is language, not a software exploit. Harari's chain of reasoning is that large-scale cooperation depends on trust, trust is maintained through institutions, and institutions use words in laws, ledgers, forms, and scriptures. “Hacking” civilization is his metaphor for mastering those symbolic systems and gaining influence through them. He also acknowledges the philosophical objection that law and religion cannot be reduced to words alone. 27:11–33:11(opens in a new tab)

• The struggle for attention could become a struggle for intimacy. Harari predicts that fluent emotional language will help AI form close relationships with people. He distinguishes describing love from feeling it and asks how children raised with AI companions might develop different expectations of attachment. He presents the consequences as unknown, not as a demonstrated developmental outcome. 36:49–41:06(opens in a new tab)

• AI-generated language also raises questions about self-understanding. Harari's closing concern is that machine-generated words could increasingly shape people's internal stories about themselves. His furniture analogy makes a distinction: using something machine-made is not inherently bad; retaining freedom over how to use it matters. He proposes examining thoughts rather than automatically identifying with them, a philosophical response rather than a technical safeguard. 43:50–46:52(opens in a new tab)

Full lecture: Watch on YouTube(opens in a new tab).