A2 · ElementaryAdapted from ElDiario.es

Can AI Go Out of Control?

¿Puede la IA salirse de control?

Sep 12, 20261 min readGraded Spanish · word-by-word audio

Read along

Un joven ingeniero llamado Jacob Coxon trabajó en OpenAI y Anthropic. Hace poco, dejó su trabajo y envió un mensaje al mundo: «La IA podría matarnos a todos antes de que acabe la década». No dio pruebas, pero dice que vio cosas peligrosas en esos laboratorios. Muchos expertos no creen que la IA nos mate pronto. Pero la ONU tiene un comité de 40 expertos que estudia estos riesgos. En julio publicaron un informe. Dicen que la IA avanza más rápido que nuestra capacidad para controlarla. El informe habla de «agentes de IA»: sistemas que actúan solos para completar tareas. Cuanto más capaces son, más difícil es vigilarlos. Los expertos avisan: «No hay garantías científicas de que no vayan a incumplir las instrucciones». Piden reglas internacionales claras.

Watch out, English speaker

English speakers often say No creo que viene because English uses the indicative after 'I don't think (that) he is coming.' But Spanish requires the subjunctive after a negated belief: No creo que venga. Remember: negation of belief triggers subjunctive. English uses 'the more... the more...' with two separate clauses, but Spanish uses cuanto más... más... and never más... más... alone. Also, unlike English, Spanish does not repeat 'the' before the second comparative: Cuanto más capaces son, más difícil es vigilarlos (not más lo difícil).

Key vocabulary

ingenierom.
engineer
“Un joven ingeniero trabajó en OpenAI.”— A young engineer worked at OpenAI.
Pattern ingeniero de software software engineer
Compare técnico technician
Plural ingenieros; feminine ingeniera
💡 From Latin ingenium (talent), cognate with ingenio (wit).
peligrosoadj.
dangerous
“Dice que vio cosas peligrosas.”— He says he saw dangerous things.
Pattern es peligroso is dangerous
Compare riesgoso risky
Feminine peligrosa; plural peligrosos/peligrosas
💡 From peligro (danger), the suffix -oso means 'full of...'.
expertom.
expert
“Muchos expertos no creen que la IA nos mate pronto.”— Many experts don't believe AI will kill us soon.
Pattern experto en IA AI expert
Compare especialista specialist
Plural expertos; feminine experta
💡 From Latin expertus (experienced), cognate with experiencia.
informem.
report
“En julio publicaron un informe.”— In July they published a report.
Pattern publicar un informe publish a report
Compare reporte report
Plural informes
💡 From informar (to inform), meaning 'a document that provides information'.
avanzarv.
to advance, progress
“La IA avanza más rápido que nuestra capacidad.”— AI advances faster than our ability.
Pattern avanzar rápido advance quickly
Compare progresar progress
Present avanzo, avanzas, avanza...
💡 From avance (advance), cognate with English advance.
capacidadf.
ability, capacity
“La IA avanza más rápido que nuestra capacidad para controlarla.”— AI advances faster than our ability to control it.
Pattern capacidad para hacer algo ability to do something
Compare habilidad skill
Plural capacidades
💡 From Latin capacitas, cognate with capaz (capable).
vigilarv.
to monitor, watch over
“Cuanto más capaces son, más difícil es vigilarlos.”— The more capable they are, the harder they are to monitor.
Pattern vigilar de cerca monitor closely
Compare supervisar supervise
Present vigilo, vigilas, vigila...
💡 From Latin vigilare (to stay awake), related to vela (night watch).
garantíaf.
guarantee
“No hay garantías científicas de que no vayan a incumplir las instrucciones.”— There are no scientific guarantees that they won't disobey instructions.
Pattern no hay garantías de que there is no guarantee that
Compare seguridad security
Plural garantías
💡 From garantir (to guarantee), cognate with English guarantee.
incumplirv.
to disobey, fail to comply
“No hay garantías de que no vayan a incumplir las instrucciones.”— There are no guarantees that they won't disobey instructions.
Pattern incumplir las reglas break the rules
Compare violar violate
Present incumplo, incumples, incumple...
💡 in- (negative prefix) + cumplir (to comply), meaning 'to not comply'.

Grammar spotlight

Negative belief + subjunctive

When the main clause expresses a negative belief (such as no creer que, no pensar que), the verb in the subordinate clause must be in the subjunctive. For example: No creo que venga. (I don't believe he will come.)

“Muchos expertos no creen que la IA nos mate pronto.”— Many experts don't believe that AI will kill us soon.

⚠️ Don't use the indicative mate → mata; after a negative belief, the subjunctive is required.

cuanto más... más... comparative structure

This structure means 'the more... the more...' and uses two comparatives in parallel. Cuanto más + adjective/adverb/noun, más + result.

“Cuanto más capaces son, más difícil es vigilarlos.”— The more capable they are, the harder they are to monitor.

⚠️ Don't omit the second más, and don't use que instead of a comma.

Sentence patterns to steal

La IA podría matarnos a todos antes de que acabe la década.
AI could kill us all before the decade is over.
podría + infinitive + antes de que + subjunctiveYou can replace podría with another verb, and change the subject and verb after antes de que.
Cuanto más capaces son, más difícil es vigilarlos.
The more capable they are, the harder they are to monitor.
Cuanto más + adj./adv., más + adj./adv. + es + infinitiveYou can replace capaces with another adjective, and vigilarlos with another verb.
No hay garantías científicas de que no vayan a incumplir las instrucciones.
There are no scientific guarantees that they won't disobey instructions.
No hay garantías de que + subjunctiveYou can replace garantías with another noun, and incumplir with another verb.

UN AI Expert Committee

In 2025, the United Nations established an international expert committee on artificial intelligence, made up of 40 members, including Turing Award winner Yoshua Bengio and Nobel Peace Prize laureate Maria Ressa. They work independently in their personal capacity, aiming to document scientific consensus and disagreements about AI. The committee seeks to provide a clear, science-based overview of AI risks and opportunities.