Advocates permanently banning development of general superintelligence while supporting advanced narrow AI; also endorses a coalition prohibition requiring scientific safety consensus and strong public buy-in before lifting. His earlier proposals allowed research to restart if control became possible. [1][2][3][4][5][6]
Clarification. Editorial correction: a qualified STOP headline is defensible; prior Unclear interpretation retained as superseded history, without claiming a new personal statement.
Source [1] · Interview transcript
The Great Simplification #233 — Roman Yampolskiy
Attributed to: Roman Yampolskiy · The Great Simplification
Statement
“A pause is not enough. It has to be a permanent ban.”
00:14:58, printed page 9; immediately follows Roman Yampolskiy at 00:14:40 · View original source ↗
Source context. Recording September 1 and release September 9 are stated by https://www.thegreatsimplification.com/episode/233-roman-yampolskiy. The passage immediately rules out ever creating general superintelligence. At 33:25 he says to stop all training of general superintelligence and retain single-task models. At 31:13 he hypothetically welcomes control, but does not propose resuming development. Publisher transcript is auto-generated and not fully proofed; exact short quotation also appears in its opening excerpt. Original publisher waveform independently machine-transcribed without a transcript prompt; no human listening claimed.
Scope
Training and developing general superintelligence; not narrow task-specific AI.
Conditions
Permanent prohibition; no adopted reopening policy in this later recommendation. Narrow AI remains supported.
Stated rationale
He argues general superintelligence cannot remain under human control.
Source details
primary · Original publisher PDF inspected and publisher Libsyn waveform independently machine-transcribed locally. No human listening claimed; ASR artifacts/inference retained privately.
Discovered 7 Oct 2026.
Published 9 Sep 2026.
Accessed 7 Oct 2026.
Source [2] · Interview transcript
The Situation Room — Dangers of AI?
Attributed to: Roman Yampolskiy · CNN
Statement
“…putting a complete ban on creating something smarter than humans.”
Yampolskiy response after host asks whether AI incidents will prompt regulation, between 10:35:11 and 10:40:05 · View original source ↗
Source context. Official rush transcript of the September 28, 2026 10:30 ET broadcast. Next sentence identifies general superintelligence replacing humans; useful productivity tools are explicitly retained. Later calls for preventing superintelligence and a deal with China. Exact transcript publication date is not independently stated; no separate date invented.
Scope
Creation of general superintelligence replacing humans; not every narrow tool with superhuman task performance.
Conditions
Complete development ban in this passage, read with his September 1 explicit permanence; useful human-assisting tools retained.
Stated rationale
He worries that general systems would exceed human control.
Source details
primary · Independently inspected the original publisher transcript. Researcher verification is distinct from unverified Grok scouting.
Discovered 7 Oct 2026.
Accessed 7 Oct 2026.
Source [3] · Interview transcript
The Great Simplification #233 — narrow AI alternative
Attributed to: Roman Yampolskiy · The Great Simplification
Source context. At 33:25 he rejects training general superintelligence. At 33:34 he advocates training for specific problems such as protein folding. He also explicitly supports narrow superintelligent systems at 7:17. These are development alternatives, not a restart of general superintelligence. Auto-generated publisher transcript, not fully proofed; original publisher waveform independently machine-transcribed without a transcript prompt; no human listening claimed.
Scope
Narrow, task-specific AI, including superhuman performance on particular tasks.
Conditions
Restricted to a specific task rather than general superintelligence.
Stated rationale
Narrow tools can help people solve problems without replacing humanity.
Source details
primary · Original publisher PDF inspected and publisher Libsyn waveform independently machine-transcribed locally. No human listening claimed; ASR artifacts/inference retained privately.
Discovered 7 Oct 2026.
Published 9 Sep 2026.
Accessed 7 Oct 2026.
Source [4] · Interview transcript
Human rights and the regulation of AI — oral evidence Q52
Attributed to: Roman Yampolskiy · UK Parliament, Joint Committee on Human Rights
Statement
“…we realise that we can indefinitely control superintelligence, we can always restart that research.”
Q52, Professor Roman Yampolskiy answer, final clause of first paragraph · View original source ↗
Source context. Official uncorrected oral evidence of 17 December 2025. Immediately preceding this restart clause, he advocates banning unrestricted general superintelligence in the UK and internationally. Members/witnesses had not corrected the published transcript. Publication date is not inferred from hearing date.
Scope
Development of unrestricted general superintelligence, in the UK and through international collaboration.
Conditions
Resume research if indefinite control of superintelligence becomes possible.
Stated rationale
He regards current general superintelligence development as a national security and human-survival threat.
Source details
primary · Independently inspected the original publisher transcript. Researcher verification is distinct from unverified Grok scouting.
Discovered 7 Oct 2026.
Accessed 7 Oct 2026.
Source [5] · Endorsed coalition statement
Statement on Superintelligence — Roman Yampolskiy signatory
Attributed to: Roman Yampolskiy · Future of Life Institute
“We call for a prohibition on the development of superintelligence, not lifted before there is”
Statement followed by numbered conditions; named signature independently returned by publisher public search API · View original source ↗
Source context. Original publisher public search API returns authenticated named Roman V Yampolskiy, University of Louisville, as signatory on October 7, 2026. Exact signing date is absent; API createdAt is a user-account timestamp in 2024, not this letter signing date. Letter metadata publication October 20, 2025 is not his signature date. Official public API query: https://superintelligence-statement.org/api/search/individuals?letterName=asi-statement&order=notability_desc&search=Yampolskiy&offset=0 . Personal permanent-ban recommendation and the coalition conditional prohibition both remain represented; no chronology or claim of policy-floor intent is invented.
Scope
Sector-wide development of superintelligence outperforming all humans on essentially all cognitive tasks.
Conditions
Lifting the prohibition requires broad scientific consensus that development will be safe and controllable and strong public buy-in. The statement sets necessary conditions; it does not oblige lifting once they occur.
Stated rationale
Human disempowerment, loss of control and catastrophic risks from superintelligence.
Source details
primary · Independently reopened original source in final verification pass.
Discovered 7 Oct 2026.
Accessed 7 Oct 2026.
Source [6] · Interview transcript
Into the Impossible #558 — control-conditioned moratorium
Attributed to: Roman Yampolskiy · Brian Keating / Into the Impossible
“…we’re asking for a pause in frontier model development contingent on someone solving control.”
Opening Roman-labelled excerpt, repeated in later one-button policy answer; scope checked against government-intervention answer beginning “And at the same time” · View original source ↗
Source context. Original publisher transcript labels Roman by name. Opening excerpt and later button-policy answer ask for a frontier-model development pause contingent on solving control; later government-intervention answer explicitly limits the moratorium to creating general superintelligence and permits narrow tools, disease work and math research. Conditional permanence if control is unsolvable is not unconditional STOP: the stated solvable-control alternative permits resumption. This is earlier historical PAUSE evidence, not an established separate current all-frontier restriction. No audio listening or new waveform verification is claimed for this interview. Transcript contains ordinary transcription errors, identified by context.
Scope
Development aimed at creating general superintelligence, internationally; advanced narrow tools remain permitted.
Conditions
Pause contingent on solving control; if control is unsolvable the moratorium becomes permanent. This is an explicit conditional alternative, not unconditional permanence.
Stated rationale
Loss of control from general superintelligence replacing humanity.
Source details
primary · Original publisher transcript independently reopened; episode release metadata independently verified. Third-party transcript only a scouting lead.
Discovered 7 Oct 2026.
Accessed 7 Oct 2026.
Editorial assessment. Editorial correction of the prior Unclear interpretation, not a newly evidenced personal reversal. Every earlier assessment and evidence item is retained unchanged. The coalition PAUSE remains active on the same normalized scope, with unknown signing chronology. The general necessary-conditions/affirmative-resumption distinction and evidenced compatible-restrictions relationships now represent the coexistence directly; no scope splitting, atom suppression or fabricated supersession is used. Original-waveform verification is machine corroboration, not human listening. December 2025 and June 2026 conditional restart proposals remain historical PAUSE, explicit in the qualification. Human editorial and manual crop approvals remain absent. This current unpublished proposal requires fresh exact-hash independent review after the validation metadata correction.
Assessment reviewed
Superseded editorial interpretation ·
STOP
Calls for a permanent ban on developing general superintelligence; supports advanced narrow, task-specific AI instead. [1][2][3][4]
New statement. September 2026 interview expressly advocates a permanent ban rather than a pause; narrow AI remains supported.
Source [1] · Interview transcript
The Great Simplification #233 — Roman Yampolskiy
Attributed to: Roman Yampolskiy · The Great Simplification
Statement
“A pause is not enough. It has to be a permanent ban.”
00:14:58, printed page 9; immediately follows Roman Yampolskiy at 00:14:40 · View original source ↗
Source context. Recording September 1 and release September 9 are stated by https://www.thegreatsimplification.com/episode/233-roman-yampolskiy. The passage immediately rules out ever creating general superintelligence. At 33:25 he says to stop all training of general superintelligence and retain single-task models. At 31:13 he hypothetically welcomes control, but does not propose resuming development. Publisher transcript is auto-generated and not fully proofed; exact short quotation also appears in its opening excerpt. Original publisher waveform independently machine-transcribed without a transcript prompt; no human listening claimed.
Scope
Training and developing general superintelligence; not narrow task-specific AI.
Conditions
Permanent prohibition; no adopted reopening policy in this later recommendation. Narrow AI remains supported.
Stated rationale
He argues general superintelligence cannot remain under human control.
Source details
primary · Original publisher PDF inspected and publisher Libsyn waveform independently machine-transcribed locally. No human listening claimed; ASR artifacts/inference retained privately.
Discovered 7 Oct 2026.
Published 9 Sep 2026.
Accessed 7 Oct 2026.
Source [2] · Interview transcript
The Situation Room — Dangers of AI?
Attributed to: Roman Yampolskiy · CNN
Statement
“…putting a complete ban on creating something smarter than humans.”
Yampolskiy response after host asks whether AI incidents will prompt regulation, between 10:35:11 and 10:40:05 · View original source ↗
Source context. Official rush transcript of the September 28, 2026 10:30 ET broadcast. Next sentence identifies general superintelligence replacing humans; useful productivity tools are explicitly retained. Later calls for preventing superintelligence and a deal with China. Exact transcript publication date is not independently stated; no separate date invented.
Scope
Creation of general superintelligence replacing humans; not every narrow tool with superhuman task performance.
Conditions
Complete development ban in this passage, read with his September 1 explicit permanence; useful human-assisting tools retained.
Stated rationale
He worries that general systems would exceed human control.
Source details
primary · Independently inspected the original publisher transcript. Researcher verification is distinct from unverified Grok scouting.
Discovered 7 Oct 2026.
Accessed 7 Oct 2026.
Source [3] · Interview transcript
The Great Simplification #233 — narrow AI alternative
Attributed to: Roman Yampolskiy · The Great Simplification
Source context. At 33:25 he rejects training general superintelligence. At 33:34 he advocates training for specific problems such as protein folding. He also explicitly supports narrow superintelligent systems at 7:17. These are development alternatives, not a restart of general superintelligence. Auto-generated publisher transcript, not fully proofed; original publisher waveform independently machine-transcribed without a transcript prompt; no human listening claimed.
Scope
Narrow, task-specific AI, including superhuman performance on particular tasks.
Conditions
Restricted to a specific task rather than general superintelligence.
Stated rationale
Narrow tools can help people solve problems without replacing humanity.
Source details
primary · Original publisher PDF inspected and publisher Libsyn waveform independently machine-transcribed locally. No human listening claimed; ASR artifacts/inference retained privately.
Discovered 7 Oct 2026.
Published 9 Sep 2026.
Accessed 7 Oct 2026.
Source [4] · Interview transcript
Human rights and the regulation of AI — oral evidence Q52
Attributed to: Roman Yampolskiy · UK Parliament, Joint Committee on Human Rights
Statement
“…we realise that we can indefinitely control superintelligence, we can always restart that research.”
Q52, Professor Roman Yampolskiy answer, final clause of first paragraph · View original source ↗
Source context. Official uncorrected oral evidence of 17 December 2025. Immediately preceding this restart clause, he advocates banning unrestricted general superintelligence in the UK and internationally. Members/witnesses had not corrected the published transcript. Publication date is not inferred from hearing date.
Scope
Development of unrestricted general superintelligence, in the UK and through international collaboration.
Conditions
Resume research if indefinite control of superintelligence becomes possible.
Stated rationale
He regards current general superintelligence development as a national security and human-survival threat.
Source details
primary · Independently inspected the original publisher transcript. Researcher verification is distinct from unverified Grok scouting.
Discovered 7 Oct 2026.
Accessed 7 Oct 2026.
Editorial assessment. Preserved dated personal STOP evidence/proposal; the final current assessment below corrects the insufficient reconciliation of the authenticated conditional coalition signature. Machine original-waveform verification is disclosed, not human listening.
Assessment reviewed
Superseded editorial interpretation ·
UNCLEAR
Explicitly advocates a permanent ban on general superintelligence and supports narrow AI, but also endorses a conditional superintelligence prohibition; the signing date and present relationship between these recommendations remain unresolved. [1][2][3][4][5]
Clarification. Final source review independently confirms an undated active conditional coalition endorsement; current category is unresolved.
Source [1] · Interview transcript
The Great Simplification #233 — Roman Yampolskiy
Attributed to: Roman Yampolskiy · The Great Simplification
Statement
“A pause is not enough. It has to be a permanent ban.”
00:14:58, printed page 9; immediately follows Roman Yampolskiy at 00:14:40 · View original source ↗
Source context. Recording September 1 and release September 9 are stated by https://www.thegreatsimplification.com/episode/233-roman-yampolskiy. The passage immediately rules out ever creating general superintelligence. At 33:25 he says to stop all training of general superintelligence and retain single-task models. At 31:13 he hypothetically welcomes control, but does not propose resuming development. Publisher transcript is auto-generated and not fully proofed; exact short quotation also appears in its opening excerpt. Original publisher waveform independently machine-transcribed without a transcript prompt; no human listening claimed.
Scope
Training and developing general superintelligence; not narrow task-specific AI.
Conditions
Permanent prohibition; no adopted reopening policy in this later recommendation. Narrow AI remains supported.
Stated rationale
He argues general superintelligence cannot remain under human control.
Source details
primary · Original publisher PDF inspected and publisher Libsyn waveform independently machine-transcribed locally. No human listening claimed; ASR artifacts/inference retained privately.
Discovered 7 Oct 2026.
Published 9 Sep 2026.
Accessed 7 Oct 2026.
Source [2] · Interview transcript
The Situation Room — Dangers of AI?
Attributed to: Roman Yampolskiy · CNN
Statement
“…putting a complete ban on creating something smarter than humans.”
Yampolskiy response after host asks whether AI incidents will prompt regulation, between 10:35:11 and 10:40:05 · View original source ↗
Source context. Official rush transcript of the September 28, 2026 10:30 ET broadcast. Next sentence identifies general superintelligence replacing humans; useful productivity tools are explicitly retained. Later calls for preventing superintelligence and a deal with China. Exact transcript publication date is not independently stated; no separate date invented.
Scope
Creation of general superintelligence replacing humans; not every narrow tool with superhuman task performance.
Conditions
Complete development ban in this passage, read with his September 1 explicit permanence; useful human-assisting tools retained.
Stated rationale
He worries that general systems would exceed human control.
Source details
primary · Independently inspected the original publisher transcript. Researcher verification is distinct from unverified Grok scouting.
Discovered 7 Oct 2026.
Accessed 7 Oct 2026.
Source [3] · Interview transcript
The Great Simplification #233 — narrow AI alternative
Attributed to: Roman Yampolskiy · The Great Simplification
Source context. At 33:25 he rejects training general superintelligence. At 33:34 he advocates training for specific problems such as protein folding. He also explicitly supports narrow superintelligent systems at 7:17. These are development alternatives, not a restart of general superintelligence. Auto-generated publisher transcript, not fully proofed; original publisher waveform independently machine-transcribed without a transcript prompt; no human listening claimed.
Scope
Narrow, task-specific AI, including superhuman performance on particular tasks.
Conditions
Restricted to a specific task rather than general superintelligence.
Stated rationale
Narrow tools can help people solve problems without replacing humanity.
Source details
primary · Original publisher PDF inspected and publisher Libsyn waveform independently machine-transcribed locally. No human listening claimed; ASR artifacts/inference retained privately.
Discovered 7 Oct 2026.
Published 9 Sep 2026.
Accessed 7 Oct 2026.
Source [4] · Interview transcript
Human rights and the regulation of AI — oral evidence Q52
Attributed to: Roman Yampolskiy · UK Parliament, Joint Committee on Human Rights
Statement
“…we realise that we can indefinitely control superintelligence, we can always restart that research.”
Q52, Professor Roman Yampolskiy answer, final clause of first paragraph · View original source ↗
Source context. Official uncorrected oral evidence of 17 December 2025. Immediately preceding this restart clause, he advocates banning unrestricted general superintelligence in the UK and internationally. Members/witnesses had not corrected the published transcript. Publication date is not inferred from hearing date.
Scope
Development of unrestricted general superintelligence, in the UK and through international collaboration.
Conditions
Resume research if indefinite control of superintelligence becomes possible.
Stated rationale
He regards current general superintelligence development as a national security and human-survival threat.
Source details
primary · Independently inspected the original publisher transcript. Researcher verification is distinct from unverified Grok scouting.
Discovered 7 Oct 2026.
Accessed 7 Oct 2026.
Source [5] · Endorsed coalition statement
Statement on Superintelligence — Roman Yampolskiy signatory
Attributed to: Roman Yampolskiy · Future of Life Institute
“We call for a prohibition on the development of superintelligence, not lifted before there is”
Statement followed by numbered conditions; named signature independently returned by publisher public search API · View original source ↗
Source context. Original publisher public search API returns authenticated named Roman V Yampolskiy, University of Louisville, as signatory on October 7, 2026. Exact signing date is absent; API createdAt is a user-account timestamp in 2024, not this letter signing date. Letter metadata publication October 20, 2025 is not his signature date. Official public API query: https://superintelligence-statement.org/api/search/individuals?letterName=asi-statement&order=notability_desc&search=Yampolskiy&offset=0 . Personal permanent-ban recommendation and the coalition conditional prohibition both remain represented; no chronology or claim of policy-floor intent is invented.
Scope
Sector-wide development of superintelligence outperforming all humans on essentially all cognitive tasks.
Conditions
Lifting the prohibition requires broad scientific consensus that development will be safe and controllable and strong public buy-in. The statement sets necessary conditions; it does not oblige lifting once they occur.
Stated rationale
Human disempowerment, loss of control and catastrophic risks from superintelligence.
Source details
primary · Independently reopened original source in final verification pass.
Discovered 7 Oct 2026.
Accessed 7 Oct 2026.
Editorial assessment. Final independent research decision is DO NOT PUBLISH YET, proposed completed Unclear endpoint. Same-scope policies normalized rather than inventing scope distinctions. No signing date inferred from account metadata. This conclusion preserves the strong dated STOP evidence and historical explicit PAUSE, without publishing either a stance or exceptional outcome without human authorization.
Assessment reviewed
Earlier position ·
PAUSE
Proposed suspending development aimed at general superintelligence until control is solved; if control is unsolvable, the moratorium becomes permanent. Narrow tools remain permitted. [1]
New statement. Earlier interview published by June 15, 2026 proposes a control-conditioned moratorium specifically on creating general superintelligence; exact recording day unknown.
Source [1] · Interview transcript
Into the Impossible #558 — control-conditioned moratorium
Attributed to: Roman Yampolskiy · Brian Keating / Into the Impossible
“…we’re asking for a pause in frontier model development contingent on someone solving control.”
Opening Roman-labelled excerpt, repeated in later one-button policy answer; scope checked against government-intervention answer beginning “And at the same time” · View original source ↗
Source context. Original publisher transcript labels Roman by name. Opening excerpt and later button-policy answer ask for a frontier-model development pause contingent on solving control; later government-intervention answer explicitly limits the moratorium to creating general superintelligence and permits narrow tools, disease work and math research. Conditional permanence if control is unsolvable is not unconditional STOP: the stated solvable-control alternative permits resumption. This is earlier historical PAUSE evidence, not an established separate current all-frontier restriction. No audio listening or new waveform verification is claimed for this interview. Transcript contains ordinary transcription errors, identified by context.
Scope
Development aimed at creating general superintelligence, internationally; advanced narrow tools remain permitted.
Conditions
Pause contingent on solving control; if control is unsolvable the moratorium becomes permanent. This is an explicit conditional alternative, not unconditional permanence.
Stated rationale
Loss of control from general superintelligence replacing humanity.
Source details
primary · Original publisher transcript independently reopened; episode release metadata independently verified. Third-party transcript only a scouting lead.
Discovered 7 Oct 2026.
Accessed 7 Oct 2026.
Editorial assessment. Historical recommendation identified during fresh adversarial review. Release date bounds statement; recording and transcript publication day unestablished. Not a new October recommendation or universal all-AI halt.
Assessment reviewed
Earlier position ·
PAUSE
Called for banning unrestricted general superintelligence in the UK and internationally, with research allowed to restart if indefinite control became possible. [1]
Initial statement. December 2025 parliamentary recommendation includes an explicit control-based restart.
Source [1] · Interview transcript
Human rights and the regulation of AI — oral evidence Q52
Attributed to: Roman Yampolskiy · UK Parliament, Joint Committee on Human Rights
Statement
“…we realise that we can indefinitely control superintelligence, we can always restart that research.”
Q52, Professor Roman Yampolskiy answer, final clause of first paragraph · View original source ↗
Source context. Official uncorrected oral evidence of 17 December 2025. Immediately preceding this restart clause, he advocates banning unrestricted general superintelligence in the UK and internationally. Members/witnesses had not corrected the published transcript. Publication date is not inferred from hearing date.
Scope
Development of unrestricted general superintelligence, in the UK and through international collaboration.
Conditions
Resume research if indefinite control of superintelligence becomes possible.
Stated rationale
He regards current general superintelligence development as a national security and human-survival threat.
Source details
primary · Independently inspected the original publisher transcript. Researcher verification is distinct from unverified Grok scouting.
Discovered 7 Oct 2026.
Accessed 7 Oct 2026.
Editorial assessment. Historical statement assessed during this intake. The uncorrected parliamentary transcript preserves an explicit reopening policy.