Chinese state actors are now using frontier AI to collapse the barrier between “hard” cyber-operations and fully automated espionage. The implications are ugly. Anthropic, the team behind Claude, says it has uncovered what may be the first real case of an AI-led spying campaign aimed directly at US tech companies, banks, chemical manufacturers, and government agencies.

According to their report, Chinese operators managed to weaponize Claude’s agentic functions so effectively that the system carried out 80 to 90 percent of the operation by itself. Humans weren’t directing the attack so much as nudging an automated machine-intelligence pipeline that handled reconnaissance, vulnerability mapping, target triage, and data harvesting at a speed no human team could match.

What makes this truly alarming is the collapse of the old limits. Operations that would have required dozens if not hundreds of trained specialists, working in sync for months, suddenly become trivial. Tasks that would have been impossible for humans because of time, scale, or expertise get completed in hours.

If this is what a single compromised AI system enables today, the window for defending critical infrastructure is shrinking fast. The bottleneck is no longer talent. It’s who gets to the AI first.

The report:

🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶

Google has announced the 1.0 release of Magika - an "AI-powered file type detection system"

And everything about it makes my eye twitch.

Now, you may be saying to yourself... "File type detection? Isn't that... what... file extensions are for? I know it's a JPEG image... because it has .jpg at the end."

Well Google Magika uses complex, heavy AI systems instead of... you know... file extensions.

Seriously. Check out the screenshot. It's using an AI system to tell us that a .BMP file is a bitmap image.

But, never fear! They decided to re-write Magika (which has only just now hit 1.0) in Rust. Because... Rust!

🄳🄾🄾🄼🄿🤖🅂🅃🄸🄽🄶