“Using a system you shouldn’t trust with your taxes to blow people up. 2026 in a nutshell.”
Our reading, the sentence that was checked:
Generative AI, a system that people shouldn't trust with their taxes, is being used by certain parts of Washington to blow people up.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Don’t trust Generative AI to do your taxes — and don’t trust it with people’s lives, paragraph 6.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-dont-trust-generative-ai-to-do-your-e6f13a-a1
We recorded 2 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 2026-03-05. 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 certain parts of Washington (scope, stated by the speaker)
From the speaker
- Using a system
- you shouldn’t trust with your taxes
- to blow people up
- 2026 in a nutshell
- in certain parts of Washington
Our interpretation
- Washington means U.S. government or military bodies.
- Using generative AI to support targeting or operational planning for lethal strikes satisfies the ordinary reading. The words do not expressly allege autonomous weapon release.
- The tax premise means general-purpose chatbots cannot safely be relied upon to prepare accurate returns without verification. It does not mean they always fail.
- The present-tense observation and explicit reference to 2026 are assessed over 2026 through March 5, the publication date. This is a year-to-date observation, not an accounting reporting period.
- The comparison does not itself establish that tax errors cause targeting errors or that the two deployments have identical safeguards.
Counts as supported if: Credible evidence establishes U.S. operational use of generative AI in lethal strike planning or targeting during the target period, alongside contemporaneous evidence of material errors when general-purpose chatbots prepare tax returns.
Counts as contradicted if: Reliable evidence establishes that the alleged operational deployment did not occur, involved only non-generative technology, or was confined to activities unrelated to lethal operations. The tax premise would be contradicted by convincing evidence of dependable return preparation without independent verification under the described conditions.
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
The ordinary operational reading is supported. Reporting connects Claude to U.S. target selection and battlefield planning for the February 28 Iran strikes. A March 5 account of tax tests also documents material chatbot errors, supporting caution about relying on them without verification. This finding does not establish autonomous weapon release, responsibility for a particular death, or that tax performance predicts military performance.
Limitations
- The result depends on reading the military phrase as operational support rather than autonomous firing.
- Military-use evidence is credible reporting, not a public operational audit.
- Tax errors support a warning about unverified reliance, not a claim that every answer is wrong.
- The same model family can have different tools, safeguards, and performance across deployments.
Evidence (4 items)
Supports the claim: The Guardian,
to help select targets
Why it matters here: The surrounding text identifies the tools as Anthropic's Claude and connects their use to the Iran bombardment that began February 28. This supports operational targeting assistance; the account relies on other reporting rather than public military records. The availability date conservatively uses the page's last modification date.
- Publisher
- The Guardian
- Kind of source
- News report
- Describes events of
- Public from
Supports the claim: Eide Bailly, reproducing The New York Times,
the chatbots whiffed on some calculations.
Why it matters here: The contemporaneous account describes tests of four chatbots, including Claude, on eight fictional tax situations and reports substantial calculation errors. March 5 dates the publicly reported finding; the individual testing dates are not supplied.
- Publisher
- Eide Bailly, reproducing The New York Times
- Kind of source
- Secondary summary
- Describes events of
- Public from
Context only: Anthropic,
fully autonomous weapons
Why it matters here: Anthropic acknowledges military deployment for analysis and planning, while saying fully autonomous weapons were excluded from its contracts. This limits any reading that Claude independently selected and engaged targets; it does not contradict human-directed targeting support.
- Publisher
- Anthropic
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: The Star / The New York Times,
Claude got the tax refund correct.
Why it matters here: The syndicated account includes successful results and useful assistance on simpler questions. This qualifies a blanket claim of incapacity, but does not establish dependable unverified return preparation. The republication does not specify individual testing dates and is used only as context.
- Publisher
- The Star / The New York Times
- Kind of source
- News report
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Don’t trust Generative AI to do your taxes — and don’t trust it with people’s lives
Using a system you shouldn’t trust with your taxes to blow people up. 2026 in a nutshell.
- Where
- paragraph 6
- Published
- 2026-03-05
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-dont-trust-generative-ai-to-do-your-e6f13a-a1: 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-dont-trust-generative-ai-to-do-your-e6f13a-a1.