Paul Stenhouse: Major security issues found in major internet infrastructure tools, Pentagon reveals its AI partnerships

Some of the big internet infrastructure tools have revealed major production security issues   A lot of these are not household names, but just this week we've seen critical patches released to fix things that would, could, or have, caused real world harm.   There's a bug on Linux that allows an unprivileged local user to gain root/admin access.   Another on the CPanel server application that also allows unauthenticated remote attackers to bypass authentication and gain unauthorized administrative access. That means they could take your server down, or post malicious content on your site.    If you're a business owner with a hosting provider, I'd check in with them to make sure both their Linux backbone and WHM/CPanel software is patched – both are major players in the market.      These security issues come after Anthropic shared their incredibly power Mythos model with selected partners   One of those was Linux (see above!) It's unclear if it was the AI that found the exploit, but this is the type of thing Anthropic said it could do. OpenAI is also launching its own "Cyber" model to select security partners before a public release. It's amusing because Sam Altman openly mocked Anthropic for the move to 'create marketing' around the hush hush model.      The Pentagon has made its AI partnerships   On the list are Nvidia, Microsoft, and AWS. They'll be able to deploy their models onto classified networks for "lawful operational use". What constitutes “legal use” is what got Anthropic designated a supply chain risk – it didn't want to be used for mass surveillance or autonomous weapons. The DoD says 1.3 million personnel have used their "GenAI.mil" application, which gives access to a range of models in a classified setting.    LISTEN ABOVE  See omnystudio.com/listener for privacy information.