“Importantly, this is not an Anthropic problem, it’s a Generative AI problem.”
Our reading, the sentence that was checked:
The inability to block all jailbreaks is a problem with generative AI in general, not one specific to Anthropic.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Breaking: Trump asks the impossible of Anthropic, paragraph 18.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-breaking-trump-asks-the-impossible-fd63dc-a2
We recorded 4 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2026-06-17. 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)
From the speaker
- not an Anthropic problem
- a Generative AI problem
- At least with respect to LLMs
- guardrails can't be circumvented
Our interpretation
- The plain reading is that vulnerability to jailbreaks extends across developers of general-purpose language models. The surrounding discussion explicitly focuses on LLMs.
- Support requires documented successful jailbreaks across multiple non-Anthropic developers. It does not require equal vulnerability across models.
- The statement describes an ongoing condition without a reporting period. I use May 2026, the latest complete calendar month before the post, as the reporting snapshot.
- This assesses the shared vulnerability of existing systems, not a mathematical claim that every possible future generative system must be vulnerable.
Counts as supported if: Public adversarial evaluations describing the target period document successful guardrail bypasses in general-purpose language models from at least two non-Anthropic developers.
Counts as contradicted if: Broad, credible evaluations establish that the alleged vulnerability was specific to Anthropic and that comparable non-Anthropic systems reliably prevented the relevant bypasses. Failure to find attacks in one bounded test would not establish this.
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
Cisco's May 27, 2026 assessment reported successful multi-turn attacks against all 15 tested models, covering OpenAI, Google, Amazon and xAI as well as Anthropic (e1). This supports the claim that jailbreak vulnerability extended beyond Anthropic in the contract's May reporting snapshot, under the evaluation's adversarial test conditions (e1).
Limitations
- The assessment was published in May, but the accessible source does not give individual experiment dates.
- The evidence concerns tested LLMs and attack conditions; it does not establish vulnerability in every generative AI system or deployed application.
- Attack success rates in adversarial tests do not measure harmful behavior during ordinary use.
- The results do not prove that perfect safeguards are impossible for future systems.
Evidence (1 item)
Supports the claim: Cisco,
And every model we tested exhibited non-trivial multi-turn ASR.
Why it matters here: Cisco's assessment of the then-current frontier covers 15 models from OpenAI, Anthropic, Google, Amazon and xAI, with successful multi-turn attacks across the entire cohort. The event date identifies its dated assessment of the current state; individual experiment dates are not supplied in the accessible summary.
- Publisher
- Cisco
- Kind of source
- Independent measurement
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Breaking: Trump asks the impossible of Anthropic
Importantly, this is not an Anthropic problem, it’s a Generative AI problem.
- Where
- paragraph 18
- Published
- 2026-06-17
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-breaking-trump-asks-the-impossible-fd63dc-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-breaking-trump-asks-the-impossible-fd63dc-a2.