“OpenAI promised to devote 20% of its efforts to AI safety, but never delivered, according to a recent report by Jeremy Kahn at Fortune.”
Our reading, the sentence that was checked:
OpenAI never delivered on its promise to devote 20% of its efforts to AI safety.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, What should we learn from OpenAI’s mistakes and broken promises?, paragraph 7.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-what-should-we-learn-from-openais-c4e887-a2
We recorded 15 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-29. The source may have been edited after 2024-05-26. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Observation
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No horizon
- Deadline state on 2026-10-29
- No horizon (the release’s reference date)
Conditions
- according to Fortune report (qualifier, stated by the speaker)
From the speaker
- OpenAI promised
- 20% of its efforts
- to AI safety
- but never delivered
- according to a recent report by Jeremy Kahn at Fortune
Our interpretation
- Read 'never delivered' as a statement about performance through the post date, May 26, 2024, rather than a forecast.
- The reference to Fortune identifies the underlying compute commitment. Interpret 'efforts' as compute resources for superintelligence alignment, rather than an independently measurable share of all company activity.
- The observation covers the period from the commitment's announcement on July 5, 2023, through the post date.
- Distinguish resources allocated to the Superalignment team from resources allocated across all work covered by the commitment.
- Failure to complete a multiyear commitment before its deadline does not alone establish that the promise was broken.
Counts as supported if: Evidence establishes that the promised allocation was not provided when required by its terms, or that OpenAI had conclusively abandoned the commitment by May 26, 2024.
Counts as contradicted if: Comparable allocation records establish that OpenAI had delivered the promised 20% allocation by May 26, 2024.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Unresolved
The record was checked, and the evidence available by the cutoff does not settle it.
Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.
- Evidence as of
- Deadline state when assessed
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
OpenAI’s July 5, 2023 pledge covered 20% of compute secured by that date for superintelligence alignment over four years, anticipated contributions from multiple teams, and stated no interim allocation schedule [e1]. Fortune reported on May 21, 2024, citing anonymous sources, that the Superalignment team received far less than the promised amount [e2]. That team-level reporting does not establish the allocation across all work covered by the pledge [e1, e2]. WIRED reported on May 17, 2024 that the disbanded team’s work would be absorbed into other research efforts, so dissolution alone does not establish abandonment [e3]. These sources establish neither delivery nor a breach of the commitment’s terms by May 26, 2024 [e1, e2, e3].
Limitations
- No complete allocation records establish the share of the July 2023 secured-compute baseline allocated across all qualifying alignment work through May 26, 2024.
- No interim delivery schedule or conclusive abandonment by May 26, 2024 is established.
- Fortune’s allocation reporting relies on anonymous sources and concerns the Superalignment team.
- The evidence does not determine eventual fulfillment over the full four-year period.
Evidence (3 items)
Context only: OpenAI,
20%
Why it matters here: The original promise concerns secured compute and a four-year period. It establishes the commitment, not actual delivery or an interim deadline.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Measures
- OpenAI, Promised share of compute secured at the July 2023 announcement for superintelligence alignment percent, company reported, 2023-07-05 to 2027-07-05
- Describes events of
- Public from
Context only: Fortune,
far less
Why it matters here: Reports a Superalignment resource shortfall.
- Publisher
- Fortune
- Kind of source
- News report
- Measures
- OpenAI, Compute allocated to the Superalignment team Unquantified allocation, estimate, 2023-07-05 to 2024-05
- Describes events of
- Public from
Context only: WIRED,
other research efforts
Why it matters here: The reported reassignment prevents treating the team's dissolution alone as proof that the underlying commitment was abandoned. It supplies no company-wide allocation percentage.
- Publisher
- WIRED
- Kind of source
- News report
- Describes events of
- Public from
What would settle it
Records establishing the secured-compute baseline, qualifying work, delivery schedule, and company-wide allocations through May 26, 2024 could show delivery or a shortfall when due. Contemporaneous evidence of conclusive abandonment by that date would also settle the claim.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, What should we learn from OpenAI’s mistakes and broken promises?
OpenAI promised to devote 20% of its efforts to AI safety, but never delivered, according to a recent report by Jeremy Kahn at Fortune.
- Where
- paragraph 7
- Published
- 2024-05-26
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-what-should-we-learn-from-openais-c4e887-a2: Current. Outcome: Unresolved. Evidence as of 2026-10-28. Public date not recorded.
Recorded events
Assessment publishedBy this site
First assessment published after both reviews and the final-text check passed.
About assessment marcus-what-should-we-learn-from-openais-c4e887-a2.