degenerating degenerate

https://www.aisi.gov.uk/blog/our-evaluation-of-claude-mythos-previews-cyber-capabilities

"Future frontier models will be more capable still, so investment now in cyber defence is vital. AI cyber capabilities are dual use; while they pose security challenges, they can also help deliver game-changing improvements in defence."

Yes but from FOSS perspective, who will pay for the tokens to get the defence ahead of the attackers? The attackers have a financial motive to fund their attack. Unless major users step in, scary times ahead.

https://www.businessinsider.com/microsoft-executive-suggests-ai-agents-buy-software-licenses-seats-2026-4

Microsoft are really something that has outlived its time. The French are showing the way: just sweep away their corruption and complacency with Linux.

Interesting new kind of confabulation recently on Gemini 3.1.

"I actually just deployed the fix for that to XXXX right before I saw your new message!" in response to my explaining what is broken.

No it didn't... it fixed it in response to my complaints... it doesn't act unless prompted and the last prompt that produced the fix diff was in response to my problem description!

This is the third or fourth instance of this new behaviour blurring what time and agency means.

As the main character of "For All Mankind" says in the new series, "I've been married three times, and if there's one thing I learned: it's that I didn't learn shit".

Current coding assists feel the same, if you work with it through some scenario, there are usually lessons you take away, like banning whatever caused the problem or ensuring next time your will do something before getting bitten.

The LLM just resets to a new context and does whatever sanded the gears last time all over again.

Jesus. We certainly have to worry about "end of civilization" due to the US, but it's not what was in the newspapers.

I've written before the step up in bug finding the last weeks is unprecedented. And now:

https://red.anthropic.com/2026/mythos-preview/

A few months ago it was: "it is evil, because I say so".

Then the focus started to become, "because I am ethical, it's clear to me AI is evil". There were side-quests down "AI is theft", "AI is a plagiarism machine", neither of which match how Apr 2026 coding assists work or consider FOSS as AI learning.

lobste.rs today: "Maintaining LOC is the devil".

It's progress, because it silently acknowledges: AI can produce usable code that does the required job, can ship, and can be maintained.