Friday, May 01, 2026

Claude Mythos Preview and Project Glasswing [UPDATED]

 


April 7
 

UPDATE: What do I know?

https://www.msn.com/en-us/news/other/pentagon-keeps-anthropic-claude-ban-separates-mythos-amid-national-security-concerns/ar-AA22bd5F

https://www.msn.com/en-in/news/world/mythos-effect-trump-administration-seeks-to-end-pentagons-ai-standoff-reinstate-anthropics-claude-says-report/ar-AA21Ye2X 

Did the DOGE dogs install backdoors? The govrenment should look into that. You'd guess that might be something somebody there might consider.

Do it as a MYTHOS project. Learn the product, how it might help.

But again, what do I know? 

 FURTHER: Another thought is the Mythos Preview could enable the government to plug Steven Miller into it, or vice versa, and do bug fixing. There seems to likely be value to doing so. Like a Robotics 101 project.
 
He looks so lifelike. 
 
FURTHER: A speculation, the Anthropic approach to this capable coding tool, debugging included, seems to be an anti-Fascist thing. As in its a step where ANTIFA might approve. Give the firm an ANTIFA gold star and a head pat.
 
 FURTHER: I am unsure what it has to do with Project Glasswing, etc, but Dan ant left.mn posted:

Lots of experts want a halt to AI in schools

 It is easy to be of two minds about that. Showing students why there'll be no jobs when they leave high school seems an argument to keep it there. Again, what do I know? It's an age where risk aversion is not as strong is in (some) adulthood, little adventuresome minds, think if they got a Mythos Preview entitlement. The brightest might enjoy having a poke at it. Goes with the gaming station at home.
 
Out of curiosity, the suggestion is that Mythos can process binary - in one of the initially linked items, the embedded video, one of the guys talked about keying in some binary and gaining administrator privilege remotely, if I remember correctly. The gamers could do a new thing, debugging vulnerabilities in their favorite games; an extension of the experience.
 
FURTHER: MSN reposts a WSJ item (paywalled at WSJ) about the Mythos situation.  See also, herehere and here followup to the WSJ post.
 
MSM reposts two other paywalled original items, WSJ here, Bloomberg here
 
FURTHER: At a guess, power grids will need to be secured. If grids go down security fixes cannot be installed if the power is off. Grids have recently had to deal with community solar and wind, and in some cases more fragmented intermittent sources, with supply - demand matching more critical than in other applications with more slack. And software contending firms in the niche often are boutiques, as well as international giants such as Siemens (not on the list of the original Anthropic release list). At a guess Siemens will not be left out. Too big to fail.
 
But thinking of Stuxnet and Iranian misbehaving Uranaium centrifuges, and the targeted software getting out into the wild, it is not pop the popcorn and sit back and watch. It is going to be far uglier than that, even with in this instance Anthropic being the good citizen the Glasswing effort suggests. 
 
Suppose Iranian and Russian drone programming vulnerabilities are uncovered. Is the US going to fix them, or allow it of firms they govern? Bet that and lose. The spooks will power some things, and how that shakes out ultimately can, as often, involve blowback. But they are as they are, and do as they do. 
 
 FURTHER: In a world of globalization, how might Trump and his administration handle international software vulnerabilities? Charlie Parker, and leave it there.