“Nobody knows for SURE that there are no serious possible consequences of open source AI, e.g in terms of bioweapons, misinformation generation etc, as Séb Krier at DeepMind nicely highlights above.”
Our reading, the sentence that was checked:
No one knows for certain that open source AI has no serious possible consequences, such as for bioweapons or misinformation generation.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Where we are right now on open source and potential AI risk., paragraph 1.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-where-we-are-right-now-on-open-source-1c7741-a2
We recorded 4 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 2023-11-03. 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
- e.g. bioweapons, misinformation generation (qualifier, stated by the speaker)
From the speaker
- Nobody knows for SURE
- no serious possible consequences of open source AI
- e.g in terms of bioweapons, misinformation generation
Our interpretation
- The ordinary reading is that publicly established scientific knowledge did not conclusively exclude serious harms from openly released AI. This is not a claim about every person's private beliefs.
- Bioweapons and misinformation are examples, not predictions that either harm must occur.
- Open source includes the public release of model weights discussed in this debate.
- The claim describes the state of knowledge at publication. With no reporting period specified, the target period is the current calendar year through the post, covering the contemporaneous research informing that state.
Counts as supported if: Contemporaneous research identifies unresolved, credible pathways to serious misuse of openly released models, inconsistent with a conclusive assurance of harmlessness.
Counts as contradicted if: Contemporaneous scientific evidence conclusively excludes serious consequences across the relevant uses and modifications of openly released AI. Favorable results on limited tests are insufficient.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Supported
The dated evidence matches what the claim committed to, under the criteria frozen before grading.
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
Research and expert discussion available before November 3, 2023 support the claim that serious possible harms from open-weight AI had not been conclusively ruled out [e1, e2]. The October 31 preprint reported that fine-tuning public Llama 2-Chat weights could undo safety training and highlighted uncertainty about safeguards when attackers have model weights [e1]. Stanford's account, available October 3, described potential fraud, downstream harms and benefits of open foundation models [e2]. These findings support uncertainty about possible harm; they do not establish that serious harm occurred or would occur [e1, e2].
Limitations
- The experiments concern public Llama 2-Chat weights and do not establish equivalent results for every open model or for access through a controlled service [e1].
- The sources do not quantify the probability or severity of actual bioweapon or misinformation harms [e1, e2].
- The evidence concerns public research and discussion, not what every individual privately knew [e1, e2].
Evidence (2 items)
Supports the claim: arXiv; Simon Lermen, Charlie Rogers-Smith and Jeffrey Ladish,
when attackers have access to model weights
Why it matters here: The researchers explicitly describe unresolved protection against misuse and report experiments undoing safety training in publicly available Llama 2 models. This supports the absence of an established safety guarantee, without proving that serious real-world harm occurred.
- Publisher
- arXiv; Simon Lermen, Charlie Rogers-Smith and Jeffrey Ladish
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: Stanford Institute for Human-Centered Artificial Intelligence,
not costless to release highly capable systems
Why it matters here: The organizer's account records expert discussion of downstream risks, security vulnerabilities and benefits of open foundation models. It corroborates an unsettled risk assessment rather than a consensus that serious consequences were impossible.
- Publisher
- Stanford Institute for Human-Centered Artificial Intelligence
- Kind of source
- Secondary summary
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Where we are right now on open source and potential AI risk.
Nobody knows for SURE that there are no serious possible consequences of open source AI, e.g in terms of bioweapons, misinformation generation etc, as Séb Krier at DeepMind nicely highlights above.
- Where
- paragraph 1
- Published
- 2023-11-03
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-where-we-are-right-now-on-open-source-1c7741-a2: Current. Outcome: Supported. 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-where-we-are-right-now-on-open-source-1c7741-a2.