Supports capability-linked safety benchmarks that can slow or pause development; concrete thresholds remain to be worked out, and this is his personal view. [1][2]
Clarification. The complete personal answer supports an advocated progression constraint even though the policy is not fully specified or implemented. It does not declare a current halt. The July shared tools letter is preparation-only and adds no category. The newer FT interview preview is directionally compatible but full original remains inaccessible and is not treated as verified evidence.
Source [1] · Original statement
Shane Legg interview
Attributed to: Shane Legg · Dwarkesh Patel
Published
“I think that's a sensible thing to do but it's actually quite hard to do. There are some people thinking about that. I know Anthropic has put out some things like that. We were thinking about similar things but actually putting concrete things down is quite a hard thing to do. I think it's an important problem and I certainly encourage people to work on it.”
0:29:38–0:29:58; full answer to capability-linked benchmark/pause/slowdown question at 0:29:26 · View original source ↗
Source context. Immediately preceding question explicitly asks whether capability levels should trigger safety benchmarks and a pause or slowdown. Legg calls this sensible and important; earlier clarifies the safety-policy answer is his personal take.
Scope
Personal view on development of increasingly capable AI; explicitly distinguished from an official Google DeepMind position at 0:29:11.
Conditions
Concrete thresholds are difficult to specify; he endorses the framework and encourages work on its implementation rather than declaring a presently closed gate.
Stated rationale
The preceding safety discussion seeks reliable ethical reasoning and oversight as capabilities grow.
Source details
primary · Publication date on original publisher transcript; recording day not independently established. Do not mistake July signatory page’s month for an exact signature date.
Discovered 3 Oct 2026.
Published 26 Oct 2023.
Accessed 3 Oct 2026.
Source [2] · Original statement
Shane Legg interview
Attributed to: Shane Legg · Dwarkesh Patel
Published
“I think that's a sensible thing to do but it's actually quite hard to do.”
0:29:38; answer to 0:29:26 question about capability-linked safety benchmarks triggering pause or slowdown · View original source ↗
Source context. Immediately preceding question explicitly asks whether capability levels should trigger safety benchmarks and a pause or slowdown. Legg calls this sensible and important; earlier clarifies the safety-policy answer is his personal take.
Scope
Development of increasingly capable AI with capability-linked safety benchmarks.
Conditions
Concrete thresholds are difficult to specify; support is for building workable conditional mechanisms.
Stated rationale
Interpret only the verified passage in its stated scope; do not infer pace from identity, optimism, risk concern or deployment policy.
Source details
primary · Publication date on original publisher transcript; recording day not independently established. Do not mistake July signatory page’s month for an exact signature date.
Discovered 3 Oct 2026.
Published 26 Oct 2023.
Accessed 3 Oct 2026.
Editorial assessment. The complete personal answer supports an advocated progression constraint even though the policy is not fully specified or implemented. It does not declare a current halt. The July shared tools letter is preparation-only and adds no category. The newer FT interview preview is directionally compatible but full original remains inaccessible and is not treated as verified evidence. This local proposal preserves the earlier evidence and interpretation. Fresh independent adversarial review and exact-hash human publication approval are separate and remain pending.
Assessment reviewed
Earlier position ·
PACE
Supports capability-linked safety benchmarks that can slow or pause development; not a blanket or fixed-duration halt. [1]
Initial statement. Research draft; independent adversarial review and human editorial approval remain separate.
Source [1] · Original statement
Shane Legg interview
Attributed to: Shane Legg · Dwarkesh Patel
Published
“I think that's a sensible thing to do but it's actually quite hard to do.”
0:29:38; answer to 0:29:26 question about capability-linked safety benchmarks triggering pause or slowdown · View original source ↗
Source context. Immediately preceding question explicitly asks whether capability levels should trigger safety benchmarks and a pause or slowdown. Legg calls this sensible and important; earlier clarifies the safety-policy answer is his personal take.
Scope
Development of increasingly capable AI with capability-linked safety benchmarks.
Conditions
Concrete thresholds are difficult to specify; support is for building workable conditional mechanisms.
Stated rationale
Interpret only the verified passage in its stated scope; do not infer pace from identity, optimism, risk concern or deployment policy.
Source details
primary · Publication date on original publisher transcript; recording day not independently established. Do not mistake July signatory page’s month for an exact signature date.
Discovered 3 Oct 2026.
Published 26 Oct 2023.
Accessed 3 Oct 2026.
Editorial assessment. The July 2026 Pacing the Frontier primary letter lists him as a personal-capacity signatory. September 16 FT interview reportedly says capability must not outrun safety; original full interview not yet verified. Neither contradicts the dated original 2023 answer. Unresolved: None. No human publication approval.