“Not all current AI is dangerous. A lot of current AI is not dangerous.”
Our reading:
Much current AI, such as GPS navigation and AlphaFold, is not dangerous.
Assessment: Unresolved
What happened
Was it true for the period described?
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 sources are compatible with many AI applications being low risk, but they do not settle the claim for October 2026. AlphaFold's release assessment predates the target period, while a later study identifies limitations in a particular safety-related use. Neither establishes the prevalence of low-risk current AI. Individual risks also do not refute Marcus's broader statement.
Caveats
- The unresolved outcome reflects evidence and period gaps, not a finding that most AI is dangerous.
- Regulatory risk categories and release judgments are not comprehensive measurements of realized harm.
- The stronger contextual assertion that AlphaFold causes no harm is not substituted for the selected claim.
Evidence (2 items)
Context only: Google DeepMind,
safely released.
Why it matters: DeepMind judged AlphaFold 3 suitable for release after assessing biosecurity risks. This earlier, application-specific judgment does not establish the prevalence of low-risk AI in October 2026 or absolute harmlessness.
- Publisher
- Google DeepMind
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- About events on
- Published
Context only: arXiv,
fail to detect
Why it matters: The evaluation includes AlphaFold 3 and challenges reliance on these tools as biological threat filters. It limits a blanket safety inference but neither documents October 2026 harm nor refutes the existence of many low-risk applications.
- Publisher
- arXiv
- Kind of source
- Preprint, not peer reviewed
- About events on
- Published
How we judged it
- Kind of claim
- Observation
- Deadline
- None given. We don’t add one.
Where we had to interpret
- The ordinary contextual reading is that many narrow AI applications pose little material danger in ordinary use, rather than that they have zero conceivable risk.
- GPS navigation and AlphaFold illustrate the claim; they do not define the entire population of current AI.
- I use October 2026 through the post date as the latest reportable month for this ongoing observation. Marcus supplies no reporting period.
- A lot means a substantial, varied set of applications, without imposing an invented numerical percentage.
Right if: Representative safety assessments and deployment records establish that many current AI applications pose little material danger in ordinary use.
Wrong if: Representative evidence establishes material danger across the supposedly low-risk applications sufficient to undermine the claim that many are low risk. An isolated harmful incident does not refute the whole proposition.
We wrote this test after the outcome was known.
What would settle it
Dated, representative assessments of narrow AI applications in the target period, including their deployment exposure, safety limits and documented harms.
We found 32 claims in this post. See them all.
Full record
Source and dates
Marcus on AI, Report from the Vatican, paragraph 31.
- 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
Not all current AI is dangerous. A lot of current AI is not dangerous.
- Where
- paragraph 31
- 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.