SN 1082: The Malicious Use of AI

Security Now #1082 - “The Malicious Use of AI”

Security Now episode #1082 is now available.

  • Anthropic’s red team reveals how AI is being weaponized for cyberattacks faster than defenses can adapt
  • Critical vulnerabilities including potential Cisco SD-WAN 0-days, PHP security issues, and HTTP/2 bomb disclosures
  • Discovery of the first AI-enabled Internet worm and accelerating attack cycles
  • Real-world incidents: $20 million ransom payment by law firms and teen-targeted WeedHack spyware
  • Lessons from a year of tracking Claude abuse and emerging AI security challenges

#SecurityNow #Cybersecurity #AI #InfoSec #MaliciousAI

Watched #1082 today then find out China’s New AI Is 6X More Efficient Than Claude and can run locally. The sky has fallen. Kimi K2.7 Code and GLM-5.2

1 Like

Because the Chinese have not had access to modern, efficient NPUs and GPUs, they have had to make do with old hardware and therefore make their code more efficient. This has been the case for decades, new software is often bloated and uses all the resources they can get, developers on restricted platforms get creative and make the code more efficient.

A prime example was Goldrunner on the C64 and Amiga. I had a friend, who had a friend, who worked for the company that made it and we got to look at the code. As far as I can remember, the code had a delay loop on the C64 to slow the game down on lower levels, it implemented the loop in around 6 bytes of code. The Amiga? The programmer seemed to have discovered copy and paste, and they had oodles of memory to play with, so the 6 byte delay loop switched to 16K of NOP (no output) instructions and the code jumped further into the NOP-list further along for each level, to reduce the amount of the delay.

1 Like

Point is, the newest China models are 6x less token expensive and with only a slight percentage less efficient than our most powerful models. Add the ability to run locally without cyber security guardrails and that dooms day Steve Gibson talked about coming soon is already here.