Halts dangerous AGI and self-improvement pathways pending safe controllability and governance; continues controllable tool AI. [1][2][3][4]
Clarification. Dated evidence assessment; subsequent source clarifies scope and conditions without inferring reversal.
Source [1] · Authored essay
World leaders at the UN call to rein in superintelligence race
Attributed to: Anthony Aguirre · Anthony Aguirre / Future of Life Institute
Published
“A halt to the development of the most dangerous capabilities, in particular recursive self-improvement.”
Aguirre statement, third bullet under every country must immediately implement and enforce. · View original source ↗
Source context. Personally authored/attributed current statement. Four immediate measures are a package, not generic capacity-building. It praises pacing/coordination but requires concrete halt and controllability guarantees, consistent with the prior conditional gate closure.
Scope
Most dangerous AI development capabilities, especially recursive self-improvement leading toward uncontrollable superintelligence.
Conditions
Licensing must guarantee safe controllability before development/release; liability and market exclusion for noncompliance; preserve controllable AI tools for beneficial uses. No calendar duration supplied.
Stated rationale
Personally authored/attributed current statement. Four immediate measures are a package, not generic capacity-building. It praises pacing/coordination but requires concrete halt and controllability guarantees, consistent with the prior conditional gate closure.
Source details
primary · Original published and first-publication footer September 24, 2026. Do not assign surrounding Monday/Tuesday/Wednesday UN dates to his own statement.
Discovered 4 Oct 2026.
Published 24 Sep 2026.
Accessed 4 Oct 2026.
Source [2] · Authored essay
Keep the Future Human — arXiv version 4
Attributed to: Anthony Aguirre · Anthony Aguirre / arXiv
Published
“…closing the ”gates” to smarter-than-human, autonomous, general-purpose AI”
Executive summary, paragraph immediately before AI is different. · View original source ↗
Source context. Solo authored Anthony Aguirre. Hard compute caps, strict liability and safety tiers are described; tool AI development continues. Later executive-summary paragraph explicitly leaves eventual sovereign systems possible, preventing an unconditional STOP reading.
Scope
Smarter-than-human autonomous general-purpose AI and superintelligence; powerful controllable tool AI remains allowed.
Conditions
Close capability gates now; future more sovereign systems only after scientific understanding and governance capacity permit safety, chosen deliberately by humanity.
Stated rationale
Solo authored Anthony Aguirre. Hard compute caps, strict liability and safety tiers are described; tool AI development continues. Later executive-summary paragraph explicitly leaves eventual sovereign systems possible, preventing an unconditional STOP reading.
Source details
primary · arXiv v4 submission history establishes March 7, 2025 12:10:22 UTC publication; manuscript title page says March 5, 2025. Use publication date, not inferred public release March 5. v3 auto-rendered August 24, 2026 is not its historical publication date.
Discovered 4 Oct 2026.
Published 7 Mar 2025.
Accessed 4 Oct 2026.
Source [3] · Authored essay
Keep the Future Human — future choice qualification
Attributed to: Anthony Aguirre · Anthony Aguirre / arXiv
Published
“…only after we have developed the scientific understanding and governance capacity to do so safely.”
Executive summary, We can engineer Tool AI to empower humanity, final paragraph referring to more sovereign systems. · View original source ↗
Source context. Explicit resumption/choice condition in same authored essay; not a prediction that it must resume. This material qualification must travel with the gate closure.
Scope
Possible future more powerful sovereign AI systems beyond controllable tools.
Conditions
Scientific understanding and governance capacity to develop safely; humanity as a whole must deliberately choose, not a default corporate/national race.
Stated rationale
Explicit resumption/choice condition in same authored essay; not a prediction that it must resume. This material qualification must travel with the gate closure.
Source details
primary · Same arXiv v4 publication March 7, 2025; manuscript dated March 5, separate drafting date.
Discovered 4 Oct 2026.
Published 7 Mar 2025.
Accessed 4 Oct 2026.
Source [4] · Authored essay
Statement on Anthropic self-improvement risks
Attributed to: Anthony Aguirre · Anthony Aguirre / Future of Life Institute
Published
“…a pause or slowdown in certain developmental pathways is crucial to protect lives and livelihoods everywhere”
Direct Aguirre statement after the introduction naming FLI President and CEO. · View original source ↗
Source context. Original FLI publication personally attributes released statement to Aguirre. Reports companies recognizing need while affirming willingness to cooperate; preserve as recent clarification, not generic risk concern.
Scope
Developmental pathways risking runaway to superintelligence.
Conditions
Collaborate with companies recognizing need for pause/slowdown; not every AI application; no fixed duration supplied.
Stated rationale
Original FLI publication personally attributes released statement to Aguirre. Reports companies recognizing need while affirming willingness to cooperate; preserve as recent clarification, not generic risk concern.
Source details
primary · Original publication and first-published footer June 8, 2026. Separate release/utterance day not explicitly supplied.
Discovered 4 Oct 2026.
Published 8 Jun 2026.
Accessed 4 Oct 2026.
Editorial assessment. Private roster candidate. Original evidence supports a scoped recommendation; preserve contrary evidence and explicit conditions. No approval, portrait or public integration. Category is provisional pending independent adversarial review.
Initial statement. Initial private assessment based on original dated evidence; not public editorial approval.
Source [1] · Authored essay
Keep the Future Human — arXiv version 4
Attributed to: Anthony Aguirre · Anthony Aguirre / arXiv
Published
“…closing the ”gates” to smarter-than-human, autonomous, general-purpose AI”
Executive summary, paragraph immediately before AI is different. · View original source ↗
Source context. Solo authored Anthony Aguirre. Hard compute caps, strict liability and safety tiers are described; tool AI development continues. Later executive-summary paragraph explicitly leaves eventual sovereign systems possible, preventing an unconditional STOP reading.
Scope
Smarter-than-human autonomous general-purpose AI and superintelligence; powerful controllable tool AI remains allowed.
Conditions
Close capability gates now; future more sovereign systems only after scientific understanding and governance capacity permit safety, chosen deliberately by humanity.
Stated rationale
Solo authored Anthony Aguirre. Hard compute caps, strict liability and safety tiers are described; tool AI development continues. Later executive-summary paragraph explicitly leaves eventual sovereign systems possible, preventing an unconditional STOP reading.
Source details
primary · arXiv v4 submission history establishes March 7, 2025 12:10:22 UTC publication; manuscript title page says March 5, 2025. Use publication date, not inferred public release March 5. v3 auto-rendered August 24, 2026 is not its historical publication date.
Discovered 4 Oct 2026.
Published 7 Mar 2025.
Accessed 4 Oct 2026.
Source [2] · Authored essay
Keep the Future Human — future choice qualification
Attributed to: Anthony Aguirre · Anthony Aguirre / arXiv
Published
“…only after we have developed the scientific understanding and governance capacity to do so safely.”
Executive summary, We can engineer Tool AI to empower humanity, final paragraph referring to more sovereign systems. · View original source ↗
Source context. Explicit resumption/choice condition in same authored essay; not a prediction that it must resume. This material qualification must travel with the gate closure.
Scope
Possible future more powerful sovereign AI systems beyond controllable tools.
Conditions
Scientific understanding and governance capacity to develop safely; humanity as a whole must deliberately choose, not a default corporate/national race.
Stated rationale
Explicit resumption/choice condition in same authored essay; not a prediction that it must resume. This material qualification must travel with the gate closure.
Source details
primary · Same arXiv v4 publication March 7, 2025; manuscript dated March 5, separate drafting date.
Discovered 4 Oct 2026.
Published 7 Mar 2025.
Accessed 4 Oct 2026.
Editorial assessment. Private roster candidate. Original evidence supports a scoped recommendation; preserve contrary evidence and explicit conditions. No approval, portrait or public integration. Category is provisional pending independent adversarial review.