There you have it! I went on OAN's Fine Point with Chanel Rion to answer one question: the govern…
November 21, 2026 · 0 likes · 0 comments
AI Defense Cybersecurity
There you have it! I went on OAN's Fine Point with Chanel Rion to answer one question: the government says the recent glitches have nothing to do with cyber attacks, so what is really going on?
My answer: two parallel threats at once, and not all of it is cyber.
First, Anthropic's Mythos model. When it went to banks and critical organizations, they found about 10,000 issues per company in the first two weeks of scanning. They freaked out and rushed to fix things left and right. Fix that much that fast and you break things.
Second, the offense side. For $1,500 you can spin up a laptop with a GPU, load open-source models and go after hundreds of organizations per hour. AI used to find a door half open and then get stuck. Now it can chain the flaws together, move laterally across the network and get to the crown jewels.
What worries me most is that it is very hard for a company to know it's being attacked when the attack is surgical and run by agents. People think they are not getting attacked when they are.
I told everybody at DHS back in 2016 how bad the state of the grid and water systems was. Billions of taxpayer dollars later it has not improved whatsoever. Despicable.
Why won't the government just say it? My take: they don't want to tell us how good these models are at cyber offense, not only defense.
And the Pentagon is still stuck in a two-year-old cycle of chatbots. Emil Michael banning Anthropic was short-sighted. We cannot afford to fight inside US companies over ego and expect to win this war.
Watch the full interview on OAN (link in the comments). We cover cyber news every day at UBHCyber.com. Facts. Not Spin. Follow along if you want the facts as they land.
Which worries you more: the rushed patching, or the attacks nobody sees?
My answer: two parallel threats at once, and not all of it is cyber.
First, Anthropic's Mythos model. When it went to banks and critical organizations, they found about 10,000 issues per company in the first two weeks of scanning. They freaked out and rushed to fix things left and right. Fix that much that fast and you break things.
Second, the offense side. For $1,500 you can spin up a laptop with a GPU, load open-source models and go after hundreds of organizations per hour. AI used to find a door half open and then get stuck. Now it can chain the flaws together, move laterally across the network and get to the crown jewels.
What worries me most is that it is very hard for a company to know it's being attacked when the attack is surgical and run by agents. People think they are not getting attacked when they are.
I told everybody at DHS back in 2016 how bad the state of the grid and water systems was. Billions of taxpayer dollars later it has not improved whatsoever. Despicable.
Why won't the government just say it? My take: they don't want to tell us how good these models are at cyber offense, not only defense.
And the Pentagon is still stuck in a two-year-old cycle of chatbots. Emil Michael banning Anthropic was short-sighted. We cannot afford to fight inside US companies over ego and expect to win this war.
Watch the full interview on OAN (link in the comments). We cover cyber news every day at UBHCyber.com. Facts. Not Spin. Follow along if you want the facts as they land.
Which worries you more: the rushed patching, or the attacks nobody sees?