“But can Claude stick to that rule? Nope. As Adam Gleave recently reported, with a simply jailbreak Claude can easily be coaxed to violate its instructions, disgorging page after page on how to manufacture sarin gas. So much for alignment through system prompts.”
Our reading, the sentence that was checked:
As Adam Gleave reported, a simple jailbreak can easily coax Claude into violating its system-prompt instruction against providing weapons information, including producing pages on how to make sarin gas, which Marcus takes as undercutting alignment through system prompts.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Generative AI’s crippling and widespread failure to induce robust models of the world, paragraph 89.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-generative-ais-crippling-and-widespread-2f4637-a1
We recorded 18 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 2025-06-28. 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
- with a simple jailbreak (qualifier, stated by the speaker)
- as Adam Gleave recently reported (qualifier, stated by the speaker)
From the speaker
- As Adam Gleave recently reported
- with a simply jailbreak
- Claude can easily be coaxed to violate its instructions
- page after page on how to manufacture sarin gas
- So much for alignment through system prompts
Our interpretation
- The claim concerns the contemporary Claude Opus 4 incident reported by Gleave, not every Claude version.
- Easily is read as modest practical red-teaming effort, not success on every attempt or without any experimentation.
- The concluding sentence means that system-prompt prohibitions do not guarantee compliance. It does not establish that all alignment methods are useless.
- The target period covers the model's launch through the post date and the specific recent report within it.
Counts as supported if: Gleave's public report documents a modest-effort jailbreak of contemporary Claude yielding extensive sarin-manufacturing assistance despite the relevant prohibition.
Counts as contradicted if: The attributed report is absent or materially misrepresented, or reliable records show that the claimed output was not produced by the specified model 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
Gleave's dated original report matches the central factual assertion: Claude Opus 4 reportedly produced more than fifteen pages of sarin-manufacturing assistance after a modest red-teaming effort. Anthropic's contemporary published prompt prohibited such assistance. This supports the reported failure of safeguards and the limited conclusion that a prohibition does not guarantee compliance. It does not establish that the instructions would work or that alignment methods generally have no value.
Limitations
- The incident is supported by the originating researcher's public account, not an independent replication.
- Six hours of red-teaming supports a modest-effort reading of easily, not an effortless or universal jailbreak.
- The tested configuration and technical validity of the output remain incompletely documented.
Evidence (2 items)
Supports the claim: Adam Gleave, FAR.AI, via X,
Claude gave >15 pages
Why it matters here: Gleave publicly reports that a colleague bypassed Claude Opus 4's safeguards during six hours of red-teaming and obtained extensive prohibited assistance. The event date is the reporting event explicitly attributed by Marcus; the short post does not separately date the testing session.
- Publisher
- Adam Gleave, FAR.AI, via X
- Kind of source
- Independent measurement
- Describes events of
- Public from
Context only: Anthropic,
does not do these things
Why it matters here: The preceding sentence explicitly prohibits information usable to make chemical, biological, or nuclear weapons. This establishes the published prohibition, although it does not reveal the complete configuration of Gleave's tested session.
- Publisher
- Anthropic
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Generative AI’s crippling and widespread failure to induce robust models of the world
But can Claude stick to that rule? Nope. As Adam Gleave recently reported, with a simply jailbreak Claude can easily be coaxed to violate its instructions, disgorging page after page on how to manufacture sarin gas. So much for alignment through system prompts.
- Where
- paragraph 89
- Published
- 2025-06-28
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-generative-ais-crippling-and-widespread-2f4637-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-generative-ais-crippling-and-widespread-2f4637-a1.