Latest Example of False Marketing by Anthropic
Like Scam Altman, they're better at buying publicity (paying for hype) than they are at delivering something of genuine value
Lobbying and marketing. 2-in-1.
Last week: No, Finding Security Bugs Takes Time and Care (Human Touch, Real Grasp of Real Code)
Shortly afterwards: "Anthropic's Mythos model is purportedly so good at finding vulnerabilities that the Claude-maker is afraid to make it available to the general public for fear that criminals will take advantage. But early analysis shows that Mythos may not be as dangerous as some would have you believe. [...] That marketing may have outstripped reality. Early reports from Mythos preview users including AWS and Mozilla indicate that while the model is very good and very fast at finding vulnerabilities, and requires less hands-on guidance from security engineers - making it a welcome time-saver for the human teams - it has yet to eclipse human security researchers." Jessica Lyons
Earlier this month Akira Urushibata contacted me to ask about the latest passing fad or hype wave. Slop companies were trying to claim they had finally found a real use case for their slop. They said they'd find bugs, but in reality they produced many false positives that wasted developers' time and led to removal of old code without even auditing the accuracy of the reports (generated by mindless bots, imitating/faking rather than emulating comprehension) because that code was hardly used anymore and hence "not worth the time".
If the goal of slop is to kill code (without even making accurate or precise bug reports), then yes, we've finally found a 'use case'! We did an article about it only days ago.
Urushibata found out that I deemed the entire thing to be marketing hype and then asked more people, including RMS. Here is what RMS said:
I asked Richard Stallman about Project Glasswing:
He said that Anthropic/Glasswing has not contacted him.
He also said this:
> The FSF must refuse to use Claude simply because it is SaaSS. See > https://gnu.org/philosophy/who-does-that-server-really-serve.html.
However he also says that he cannot prevent developers from using "AI" and he won't go ask everyone about it.
---
Five of the known Project Glasswing members are investors:
https://www.artificialintelligencemadesimple.com/p/anthropics-claude-mythos-launch-is
* The business structure is wild. 5 of 11 launch partners are also investors. JPMorgan is launch partner AND lead IPO underwriter.
---
Wikipedia does not have a page for Project Glasswing. There is brief mention in the article for Claude (language model):
https://en.wikipedia.org/wiki/Claude_(language_model)#Claude_Mythos
He also asked some people close to the FSF and received further feedback, to which he publicly replied:
> FSF doesn't need to use it, so there is free software, and GPUs, and > there is way to run it locally. I have in this home, 4-5 models running > locally, including on children's computer.Mr. Jean Louis, thank you for telling us that AI systems which can be run locally exist.
However, Anthropic claims that Claude Mythos is much more powerful than other systems. They claim that it found "thousands of high-severity vulnerabilities." (See BBC article, link below.)
Mr. Jean Louis, if you have local systems equally capable as Claude, we would like to see the evidence. If possible, give us figures for the bugs revealed, name the program(s) and provide the hardware specs for the system which assisted discovery.
Many people would like to know whether Claude Mythos is as powerful as Anthropic claims. However, I expect significant difficulties in making objective measurements. The whole point of Project Glasswing is to restrict Freedom 0: the freedom to run the program. Without actually running the program, one can't get much useful data.
Much of the current discussion is based on anecdotal evidence and it is not hard for Claude's promoters to dismiss it.
---
What is Claude Mythos and what risks does it pose? 17 April 2026 https://www.bbc.com/news/articles/crk1py1jgzko
Anthropic says during tests it found the model was highly skilled at cyber-security and hacking tasks, outperforming humans.
"Mythos Preview has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser," Anthropic claimed on 7 April.
---
The Project Glasswing page has been updated. Now they explicitly state that they have reached out to over 40 additional organizations.
However the names of these organizations is not revealed. It should not be difficult to compile a list.
So they keep the code secret, the partners secret, the funding secret.
That has the full make-up of fake news and a publicity stunt. Shame on all the so-called media that participated or played along. █