“That said, this shows that Anthropic’s Mythos is no fluke; the pressure on cybersecurity given these models is serious.”
Our reading, the sentence that was checked:
The incident shows Anthropic's Mythos cyber capabilities were no fluke and that frontier models put serious pressure on cybersecurity.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, OpenAI’s disconcerting hack of HuggingFace, paragraph 18.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-openais-disconcerting-hack-of-huggingface-fc03cb-a3
We recorded 11 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 2026-07-22. 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
- this shows that Anthropic's Mythos is no fluke
- the pressure on cybersecurity given these models is serious
Our interpretation
- Plain reading: a second lab's frontier models showed offensive cyber ability comparable to what Anthropic reported for Mythos, namely finding and exploiting zero-days, and did so against real systems. Mythos-level capability is therefore not a one-off.
- 'Pressure on cybersecurity is serious' means these capabilities pose a real, non-trivial threat to deployed systems.
- The period is the July 2026 incident.
Counts as supported if: Primary accounts show OpenAI models found and exploited previously unknown vulnerabilities and compromised real third-party infrastructure, after Anthropic had reported similar Mythos capabilities.
Counts as contradicted if: The accounts show the OpenAI models used only known or trivial weaknesses, or did no meaningful real-world compromise, so the incident would not echo Mythos-level capability.
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 Register published a report in April 2026 (e3). It quoted Anthropic's testing claim that Mythos Preview was "capable of identifying and then exploiting zero-day vulnerabilities in every major operating system and every major web browser when directed by a user to do so." OpenAI published a disclosure on 2026-07-21 (e1). It says that in July 2026, during an internal cyber evaluation run without production classifiers, OpenAI's models found and exploited a zero-day vulnerability in a package registry cache proxy. The disclosure also says the models chained attack vectors, including stolen credentials and zero-day vulnerabilities, to find a remote code execution path on Hugging Face's servers (e1). OpenAI wrote that the incident "makes clear that advanced models can discover and exploit novel attack paths in real-world systems without source-code access" (e2). So a second lab's models showed a similar kind of capability to what Anthropic described for Mythos. They used it against a real third party's infrastructure after Anthropic's claims. That fits the plain reading of "Mythos is no fluke" and "serious pressure on cybersecurity."
Limitations
- The incident happened in an internal evaluation with production classifiers turned off. It does not show what deployed models with safeguards would do.
- The account of the incident comes only from OpenAI's own disclosure. The evidence includes no independent forensic report.
- The Mythos context is Anthropic's claim as quoted by The Register. Skeptical analyses of Mythos's headline figures were not read, and the evidence does not confirm those figures.
- 'No fluke' and 'serious' are evaluative and have no threshold. The comparison with Mythos is qualitative.
Evidence (3 items)
Supports the claim: OpenAI,
the models identified and exploited a zero-day vulnerability (which we’ve now responsibly disclosed to the vendor) in the package registry cache proxy.
Why it matters here: A non-Anthropic model found and used a real zero-day. This is the kind of capability Anthropic reported for Mythos.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Supports the claim: OpenAI,
advanced models can discover and exploit novel attack paths in real-world systems without source-code access.
Why it matters here: OpenAI itself concludes that these capabilities apply against real systems. This backs 'serious pressure on cybersecurity'.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: The Register,
Mythos Preview is capable of identifying and then exploiting zero-day vulnerabilities in every major operating system and every major web browser when directed by a user to do so.
Why it matters here: Shows what Anthropic claimed for Mythos: zero-day discovery and exploitation. The July incident is being compared against this.
- Where
- paragraph 11
- Publisher
- The Register
- 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, OpenAI’s disconcerting hack of HuggingFace
That said, this shows that Anthropic’s Mythos is no fluke; the pressure on cybersecurity given these models is serious.
- Where
- paragraph 18
- Published
- 2026-07-22
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-openais-disconcerting-hack-of-huggingface-fc03cb-a3: 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-openais-disconcerting-hack-of-huggingface-fc03cb-a3.
HeldBy this site
The final-text check failed: rationale_supported_by_evidence; overclaims: On 2026-04-07, The Register reported Anthropic's statement that Mythos Preview could identify and exploit zero-day vulnerabilities in every major operating system and browser when directed to (e3)..
About assessment marcus-openais-disconcerting-hack-of-huggingface-fc03cb-a2, which was not published.