“And in fact, I think if we don’t have the new research approaches, like what Joshua is trying to do, for example, and I think we need many such approaches, then I think we were in deep trouble, and the policy alone will not get us there.”
Our reading:
Without new research approaches we will be in deep trouble, and policy alone will not make AI safe.
Assessment: Unresolved
What happened
Did the forecast come true?
Outcome: Unresolved
We checked, and the evidence so far doesn’t settle it.
Judged on a plain reading. They named no measure, so we used the reading a typical informed reader would take. It is spelled out under “How we judged it”.
Evidence up to . Checked .
Why
The February 3, 2026 International AI Safety Report describes limitations in technical safeguards and institutional risk management, but does not test whether policy alone can control agent harms [e1]. Lu and Bengio's September 9 proposal combines technical supervision, system controls and governance; it does not demonstrate their effectiveness or necessity [e2]. Neither source establishes the conditional absence of effective new technical approaches or a sustained policy-only outcome that would settle the forecast [e1, e2].
Caveats
- The claim supplies no deadline or measurable threshold for adequate safety.
- Safeguard limitations and proposed remedies do not establish that policy alone must fail [e1, e2].
- The packet contains no independently verified policy-only regime meeting the early-resolution rule [e1, e2].
Evidence (2 items)
Context only: International AI Safety Report,
still show significant limitations.
Why it matters: The report describes limitations of technical and institutional safeguards. It does not test whether policy alone could control agent harms under the forecast's antecedent.
- Publisher
- International AI Safety Report
- Kind of source
- Secondary summary
- About events on
- Published
Context only: arXiv,
AI safety failures often arise across multiple layers.
Why it matters: Lu and Bengio propose combining technical supervision, system controls and governance. This establishes a research proposal, not the absence of effective new approaches or the impossibility of policy-only control.
- Publisher
- arXiv
- Kind of source
- Preprint, not peer reviewed
- About events on
- Published
How we judged it
- Kind of claim
- Conditional forecast
- Deadline
- None given. We don’t add one.
Conditions
- if we don't have new research approaches (their words)
Where we had to interpret
- I read the claim as saying that policy measures cannot reliably control the harmful actions of broadly capable AI agents without materially different technical approaches.
- Policy includes enforceable restrictions and oversight, not merely voluntary promises.
- The antecedent concerns the absence of effective new approaches during the relevant future period. The existence of research proposals does not establish whether that antecedent will hold.
- No deadline or measurable standard for adequate safety is supplied.
Right if: With the absence of effective new technical approaches established, decisive evidence shows that policy measures alone cannot reliably control the identified agent harms. Individual failures of particular policies are insufficient to establish this necessity.
Wrong if: Policy measures alone demonstrably achieve sustained, reliable control of the identified agent harms, without relying on materially new technical approaches.
We wrote this test after the outcome was known.
What would settle it
A documented policy-only regime achieving sustained, independently verified control of specified agent harms without materially new technical approaches would contradict the forecast. Support would require establishing the absence of effective new approaches and showing that policy alone cannot reliably control those harms, beyond isolated implementation failures.
We found 32 claims in this post. See them all.
Full record
Source and dates
Marcus on AI, Report from the Vatican, paragraph 30.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-11-09. The post may have been edited after 2026-10-31.
Where it was said (1)
Said once in the posts we read.
: Marcus on AI, Report from the Vatican
And in fact, I think if we don’t have the new research approaches, like what Joshua is trying to do, for example, and I think we need many such approaches, then I think we were in deep trouble, and the policy alone will not get us there.
- Where
- paragraph 30
- Published
- 2026-10-31
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Unresolved. Evidence up to 2026-11-09.
What happened to this record
PublishedBy us
Published after both AI reviews and the final check passed.