Endorses a near-term ASI ban and halt until better alignment understanding; earlier personal proposals demand indefinite global shutdown. No fixed restart date; ordinary AI is not categorically banned. [1][2][3]
Clarification. Private current scope synthesis retains the strongest ban language and conditional future-safety qualification; no explicit personal reversal is claimed.
Source [1] · Written statement
MIRI’s Position on the Ban Artificial Superintelligence Act of 2026
Attributed to: Eliezer Yudkowsky · MIRI / Aaron Scher
Published
“…the near term requires a complete, flat ASI ban.”
“What we like about the Act,” sixth bullet; read with opening endorsement and final paragraph · View original source ↗
Source context. Aaron Scher’s September 23 written statement explicitly names Soares and Yudkowsky among endorsers. It endorses an ASI ban and precursor/interim pauses, but says many staff want eventually demonstrably safe ASI and closes with an alignment-conditioned halt. Do not infer either named individual personally belongs to the “many staff” subset.
Scope
ASI and precursor capabilities including automated AI R&D; separate interim pause for other advanced AI; some licensed non-ASI medical tools may be excepted.
Conditions
Jointly endorsed on behalf of MIRI, not a sole-authored personal statement. Near-term flat ASI ban; back off until better alignment understanding; indefinite horizon, no present safe method or fixed restart date.
Stated rationale
Aaron Scher’s September 23 written statement explicitly names Soares and Yudkowsky among endorsers. It endorses an ASI ban and precursor/interim pauses, but says many staff want eventually demonstrably safe ASI and closes with an alignment-conditioned halt. Do not infer either named individual personally belongs to the “many staff” subset.
Source details
primary · Exact dated written edition independently verified. Individual signing time not specified; recorded as publication date, not an invented speech/signing day. Speaker field identifies named endorser; authorship and on-behalf-of relationship are explicit.
Discovered 4 Oct 2026.
Published 23 Sep 2026.
Accessed 4 Oct 2026.
Source [2] · Written statement
Pausing AI Developments Isn’t Enough. We Need to Shut it All Down
Attributed to: Eliezer Yudkowsky · TIME / Eliezer Yudkowsky
Published
“Shut down all the large training runs.”
Policy paragraph following “Here’s what would actually need to be done” · View original source ↗
Source context. His personally authored March 29 essay demands indefinite global shutdown, rejects the six-month letter as insufficient, and leaves safe ASI in principle possible rather than asserting an unconditional eternal ban.
Scope
Large training runs and GPU clusters capable of dangerous superhuman AI; possible tightly restricted biological tools exception.
Conditions
Indefinite and worldwide, not six months; no government/military exception. Safe ASI acknowledged in principle but requires new science absent in foreseeable circumstances.
Stated rationale
His personally authored March 29 essay demands indefinite global shutdown, rejects the six-month letter as insufficient, and leaves safe ASI in principle possible rather than asserting an unconditional eternal ban.
Source details
primary · Original byline and March29 22:01 UTC date, plus original HTML publication metadata, inspected. Written publication supplies exact source date.
Discovered 4 Oct 2026.
Published 29 Mar 2023.
Accessed 4 Oct 2026.
Source [3] · Written statement
Only Law Can Prevent Extinction
Attributed to: Eliezer Yudkowsky · LessWrong / Eliezer Yudkowsky
Published
“There ought to be a law against further escalation of AGI capabilities”
Paragraph beginning “Specifically,” after “There ought to be a law!” · View original source ↗
Source context. Personally authored April 13 essay calls for lawful international prohibition of AGI capability escalation before ASI. It offers no restart plan here, but does not erase his earlier safe-in-principle qualification or the newer jointly endorsed alignment-conditioned statement.
Scope
Further AGI capability escalation short of superintelligence; international supervised compute, not all ordinary software.
Conditions
Conservative enforceable global limit; no restart proposed in this passage. Lawful state-backed enforcement; explicitly rejects individual unlawful violence.
Stated rationale
Personally authored April 13 essay calls for lawful international prohibition of AGI capability escalation before ASI. It offers no restart plan here, but does not erase his earlier safe-in-principle qualification or the newer jointly endorsed alignment-conditioned statement.
Source details
primary · Original author page/date and article body inspected; comments are separate speakers and excluded.
Discovered 4 Oct 2026.
Published 13 Apr 2026.
Accessed 4 Oct 2026.
Editorial assessment. Stricter boundary reading: suspend until stated control/alignment understanding, not an unconditional permanent end. Permanent-ban language and no-foreseeable-safety claims are preserved; latest named organizational endorsement is not sole-authored personal testimony. Later interviews with unknown event dates remain blocked private observations and cannot supply a manufactured dated assessment. Portrait absent; no public shell or approval.
Assessment reviewed
Earlier position ·
PAUSE
Aaron Scher’s September 23 written statement explicitly names Soares and Yudkowsky among endorsers. It endorses an ASI ban and precursor/interim pauses, but says many staff want eventually demonstrably safe ASI and closes with an alignment-conditioned halt. Do not infer either named individual personally belongs to the “many staff” subset. [1]
Scope distinction. Aaron Scher’s September 23 written statement explicitly names Soares and Yudkowsky among endorsers. It endorses an ASI ban and precursor/interim pauses, but says many staff want eventually demonstrably safe ASI and closes with an alignment-conditioned halt. Do not infer either named individual personally belongs to the “many staff” subset.
Source [1] · Written statement
MIRI’s Position on the Ban Artificial Superintelligence Act of 2026
Attributed to: Eliezer Yudkowsky · MIRI / Aaron Scher
Published
“…the near term requires a complete, flat ASI ban.”
“What we like about the Act,” sixth bullet; read with opening endorsement and final paragraph · View original source ↗
Source context. Aaron Scher’s September 23 written statement explicitly names Soares and Yudkowsky among endorsers. It endorses an ASI ban and precursor/interim pauses, but says many staff want eventually demonstrably safe ASI and closes with an alignment-conditioned halt. Do not infer either named individual personally belongs to the “many staff” subset.
Scope
ASI and precursor capabilities including automated AI R&D; separate interim pause for other advanced AI; some licensed non-ASI medical tools may be excepted.
Conditions
Jointly endorsed on behalf of MIRI, not a sole-authored personal statement. Near-term flat ASI ban; back off until better alignment understanding; indefinite horizon, no present safe method or fixed restart date.
Stated rationale
Aaron Scher’s September 23 written statement explicitly names Soares and Yudkowsky among endorsers. It endorses an ASI ban and precursor/interim pauses, but says many staff want eventually demonstrably safe ASI and closes with an alignment-conditioned halt. Do not infer either named individual personally belongs to the “many staff” subset.
Source details
primary · Exact dated written edition independently verified. Individual signing time not specified; recorded as publication date, not an invented speech/signing day. Speaker field identifies named endorser; authorship and on-behalf-of relationship are explicit.
Discovered 4 Oct 2026.
Published 23 Sep 2026.
Accessed 4 Oct 2026.
Editorial assessment. Private researcher proposal; independent review and human approval still required. Scope/time limitations retained.
Assessment reviewed
Earlier position ·
STOP
Personally authored April 13 essay calls for lawful international prohibition of AGI capability escalation before ASI. It offers no restart plan here, but does not erase his earlier safe-in-principle qualification or the newer jointly endorsed alignment-conditioned statement. [1]
Scope distinction. Personally authored April 13 essay calls for lawful international prohibition of AGI capability escalation before ASI. It offers no restart plan here, but does not erase his earlier safe-in-principle qualification or the newer jointly endorsed alignment-conditioned statement.
Source [1] · Written statement
Only Law Can Prevent Extinction
Attributed to: Eliezer Yudkowsky · LessWrong / Eliezer Yudkowsky
Published
“There ought to be a law against further escalation of AGI capabilities”
Paragraph beginning “Specifically,” after “There ought to be a law!” · View original source ↗
Source context. Personally authored April 13 essay calls for lawful international prohibition of AGI capability escalation before ASI. It offers no restart plan here, but does not erase his earlier safe-in-principle qualification or the newer jointly endorsed alignment-conditioned statement.
Scope
Further AGI capability escalation short of superintelligence; international supervised compute, not all ordinary software.
Conditions
Conservative enforceable global limit; no restart proposed in this passage. Lawful state-backed enforcement; explicitly rejects individual unlawful violence.
Stated rationale
Personally authored April 13 essay calls for lawful international prohibition of AGI capability escalation before ASI. It offers no restart plan here, but does not erase his earlier safe-in-principle qualification or the newer jointly endorsed alignment-conditioned statement.
Source details
primary · Original author page/date and article body inspected; comments are separate speakers and excluded.
Discovered 4 Oct 2026.
Published 13 Apr 2026.
Accessed 4 Oct 2026.
Editorial assessment. Private researcher proposal; independent review and human approval still required. Scope/time limitations retained.
Assessment reviewed
Earlier position ·
PAUSE
His personally authored March 29 essay demands indefinite global shutdown, rejects the six-month letter as insufficient, and leaves safe ASI in principle possible rather than asserting an unconditional eternal ban. [1]
Initial statement. His personally authored March 29 essay demands indefinite global shutdown, rejects the six-month letter as insufficient, and leaves safe ASI in principle possible rather than asserting an unconditional eternal ban.
Source [1] · Written statement
Pausing AI Developments Isn’t Enough. We Need to Shut it All Down
Attributed to: Eliezer Yudkowsky · TIME / Eliezer Yudkowsky
Published
“Shut down all the large training runs.”
Policy paragraph following “Here’s what would actually need to be done” · View original source ↗
Source context. His personally authored March 29 essay demands indefinite global shutdown, rejects the six-month letter as insufficient, and leaves safe ASI in principle possible rather than asserting an unconditional eternal ban.
Scope
Large training runs and GPU clusters capable of dangerous superhuman AI; possible tightly restricted biological tools exception.
Conditions
Indefinite and worldwide, not six months; no government/military exception. Safe ASI acknowledged in principle but requires new science absent in foreseeable circumstances.
Stated rationale
His personally authored March 29 essay demands indefinite global shutdown, rejects the six-month letter as insufficient, and leaves safe ASI in principle possible rather than asserting an unconditional eternal ban.
Source details
primary · Original byline and March29 22:01 UTC date, plus original HTML publication metadata, inspected. Written publication supplies exact source date.
Discovered 4 Oct 2026.
Published 29 Mar 2023.
Accessed 4 Oct 2026.
Editorial assessment. Private researcher proposal; independent review and human approval still required. Scope/time limitations retained.