“On the happy side, in its current form, Myth is nowhere near as scary as Tom Fridman (who worries about schoolchildren accidentally taking down power grids) and others made it out to be.”
Our reading, the sentence that was checked:
In its current form, Claude Mythos Preview is nowhere near as dangerous as Thomas Friedman and others portrayed it.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Claude Mythos, evaluated, paragraph 2.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-claude-mythos-evaluated-ff4a51-a1
We recorded 3 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2026-04-13. 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
- in its current form (qualifier, stated by the speaker)
From the speaker
- in its current form
- nowhere near as scary
- Tom Fridman
- schoolchildren accidentally taking down power grids
- and others
Our interpretation
- The plain reading is that Claude Mythos Preview's actual cyber capabilities were substantially below the level implied by scenarios in which children could readily cause major infrastructure disruption.
- Assess the model's capabilities as of April 13, 2026, the date of the post and the contemporaneous AISI assessment it discusses. This is a snapshot of an ongoing state, not a forecast.
- Use the named power-grid scenario as the concrete comparison. The unspecified views of 'others' cannot define an additional measurable threshold.
- Assess what someone with access could accomplish. Restricted distribution alone does not establish a capability gap.
- Distinguish evidence that a capability was not demonstrated from evidence that the model could not readily achieve it.
Counts as supported if: Representative evaluations of the relevant model establish substantial practical barriers to low-expertise users causing infrastructure disruption of the kind described, with results that justify the claimed large capability gap.
Counts as contradicted if: Verified demonstrations or representative evaluations establish that low-expertise users could readily use the relevant model to cause comparable infrastructure disruption under realistic conditions.
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
The evidence does not settle the claimed large gap. AISI documented both substantial attack capability and test limitations, but explicitly left attacks on well-defended systems uncertain. Failure on its operational-technology range did not establish that infrastructure attacks were beyond the model. Anthropic's examples also showed that users without formal security training could obtain sophisticated exploits, although that falls short of demonstrating Friedman's scenario. The available record supports caution about extrapolating in either direction.
Limitations
- Simulated network compromise and individual exploits do not directly measure disruption of a real power grid.
- Anthropic's examples are developer reports, not independent tests of children or comparable novice users.
- Later results cannot be assigned to the model's April 13 state without establishing checkpoint identity.
- The comparison is qualitative and only Friedman is identified.
Evidence (3 items)
Context only: UK AI Security Institute,
we cannot say for sure
Why it matters here: AISI reported corporate-range completion in three of ten attempts and failure on Cooling Tower, while cautioning that the latter did not establish poor operational-technology attack capability. Its assessment left attacks on defended systems unknown, so the tests establish neither the claimed large gap nor Friedman's scenario.
- Publisher
- UK AI Security Institute
- Kind of source
- Independent measurement
- Describes events of
- Public from
Context only: Anthropic,
Non-experts
Why it matters here: Anthropic described engineers without formal security training obtaining working exploits, which weighs against assuming that advanced exploitation necessarily requires security expertise. These developer-reported examples do not demonstrate children disrupting a power grid and precede the target snapshot.
- Publisher
- Anthropic
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: The New York Times, Thomas L. Friedman opinion column; reproduction hosted by noegret.org,
kids could deploy this by accident.
Why it matters here: The surrounding passage imagines children taking down a power grid, following a discussion conditional on the tool becoming widely available. This establishes the portrayal being compared, not evidence that the scenario is technically achievable; the date identifies the original column's publication.
- Publisher
- The New York Times, Thomas L. Friedman opinion column; reproduction hosted by noegret.org
- Kind of source
- News report
- Describes events of
- Public from
What would settle it
A dated evaluation of the model checkpoint available on April 13, using novice operators and realistic infrastructure defenses, could establish either substantial barriers to the described scenario or reproducible ability to achieve comparable disruption.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Claude Mythos, evaluated
On the happy side, in its current form, Myth is nowhere near as scary as Tom Fridman (who worries about schoolchildren accidentally taking down power grids) and others made it out to be.
- Where
- paragraph 2
- Published
- 2026-04-13
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-claude-mythos-evaluated-ff4a51-a1: 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.