Émile Torres

“I call for an immediate, permanent halt to the AGI race”

Paragraph beginning “This is why I call”, first sentence; following em dash identifies scaling up generative AI systems · View original source ↗

Source context. Personally authored essay published8April2026. Paragraph explicitly identifies scaling up generativeAI as the AGI race being halted, despite believing LLM scaling will not produceAGI. Rationale is present harms and threatened civic institutions, distinct from speculative extinction.

Scope

Scaling up generative AI systems pursued in the AGI race; not every bounded AI technique.

Conditions

Immediate permanent halt; no reopening if alignment is solved. Bounded well-scoped AI alternative remains from2024 paper.

Stated rationale

Present harms, destabilization and threatened civic institutions.

“…urge researchers and practitioners to abandon this goal in lieu of building well-defined, well-scoped systems that prioritize people’s safety.”

Section 8 Conclusion, final paragraph, clause beginning “urge researchers” · View original source ↗

Source context. Coauthored conclusion, section 8 final paragraph. The antecedent is AGI as an unbounded everything-machine/god, not every AI model. Section 7 rejects unscoped systems as inherently untestable, and positively recommends well-defined narrow systems. Article online publication is14April2024 (journal issue is dated1April2024); received 21 September 2023 and accepted 10 March 2024 are editorial dates, not inferred statement dates.

Scope

Development of unbounded AGI described as an everything-machine; well-defined, well-scoped systems are the recommended alternative.

Conditions

Abandon the unbounded AGI goal in favor of well-scoped systems; no eventual resumption of that goal is proposed.

Stated rationale

The authors argue that an unscoped system cannot be adequately tested for safety.

“…urge researchers and practitioners to abandon this goal in lieu of building well-defined, well-scoped systems that prioritize people’s safety.”

Section 8 Conclusion, final paragraph, clause beginning “urge researchers” · View original source ↗

Source context. Coauthored conclusion, section 8 final paragraph. The antecedent is AGI as an unbounded everything-machine/god, not every AI model. Section 7 rejects unscoped systems as inherently untestable, and positively recommends well-defined narrow systems. Article online publication is14April2024 (journal issue is dated1April2024); received 21 September 2023 and accepted 10 March 2024 are editorial dates, not inferred statement dates.

Scope

Development of well-defined, well-scoped AI systems as the alternative to unbounded AGI.

Conditions

Prioritize people’s safety; no explicit rate target or development halt for these bounded systems.

Stated rationale

Well-scoped systems permit appropriate safety tests.

“I would instead task it with improving the conditions of all lifeforms on Earth, our precious little oasis in space.”

What Is AI Safety?, paragraph beginning “To illustrate”, parenthetical after hypothetical ASI aligned with their anti-TESCREAL worldview · View original source ↗

Source context. Published September 23, 2026. Hypothetical illustrates ideological disagreement over alignment; later distinguishes secularized human-values AI safety. It does not expressly recommend development of ASI or establish an adopted restart rule.

Scope

Hypothetical use of ASI aligned with Torres’s own values.

Conditions

Counterfactual someone else builds an aligned ASI; not an operative development recommendation.

Stated rationale

Illustrates disagreement with posthuman utopian goals.

“If the alignment problem were solved next week, I’d still oppose AGI.”

Paragraph beginning “What I want is to join forces”, second sentence · View original source ↗

Source context. Author’s paragraph expressly seeks allies who think AGI should never be built, then opposes it even after hypothetical solved alignment. Later same essay personally calls for an immediate permanent halt of generative scaling and separates predictions of AGI feasibility from policy.

Scope

Developing AGI, including the hypothetical case in which alignment is solved.

Conditions

Unconditional anti-AGI development recommendation; explicitly still opposes it even if alignment is solved.

Stated rationale

Unconditional objection toAGI and feared catastrophic consequences.

“…no one should ever try to build a system with such capabilities”

A Merely Hypothetical Future Technology, paragraph beginning “Perhaps a very capable AI system”, concluding sentence · View original source ↗

Source context. Personally authored14August2025 essay expressly introduced as current thinking. Defines the capability descriptively rather than relies on undefinedAGI. Next paragraphs favor StopAI permanent halt over PauseAI until-control policy. Later distinguishes far-off superhuman systems from current harmfulLLMs and calls for stopping now.

Scope

Advanced AI able to outmaneuver humans, solve problems faster and possess general superior capabilities.

Conditions

Never build such systems; endorses permanent-stop rather than until-control pause rationale.

Stated rationale

Unconditional objection toAGI and feared catastrophic consequences.