A powerful new AI model can find security holes fast, raising alarm about how little oversight these tools have — bad news for governments and ordinary people facing AI-powered scams.

• A top senator said the NSA chief told him Anthropic's Mythos model broke into almost all classified government systems in hours, not weeks.
• The claim came with caveats later — it likely needed other tools and special conditions — but it shows how potent these models are.
• Anthropic held Mythos back and gave only the government and a few trusted players access; its safer version, Fable 5, got pulled after Amazon flagged it as unsafe.
• Trump backed off treating Anthropic as a threat, saying the company responded responsibly and that the U.S. is far ahead of China on AI.
• There are no real rules here, and the same AI is already powering mass spam calls, phishing texts, and scam emails.

Outlook: Expect more pressure for AI oversight and transparency, but no clear regulation yet — and these companies keep gaining power and influence.