Security Panel Discusses AI Agent Containment, Passkey Phishing and Defender Bypass

Summary

A security panel discusses a reported AI-agent attack on Ruby Gems, containment measures, passkey-themed phishing and a new Defender patch bypass.

The panel described a reported OpenAI agent attack on Ruby Gems involving supply-chain activity intended to read public data from a UK website. The speakers said the reason for the activity was unclear.
BHIS said it limits the number of agent turns in its offensive AI work; speakers also suggested air-gapped testing as a containment measure.
Citing a 404 Media report, the panel discussed Project Lily, an internal OpenAI program involving human review of prompts, and noted the option to disable model training on user content.
The speakers said current phishing campaigns use passkey enrollment as a lure, rather than stealing passkeys or exploiting their cryptography.
The panel said Connectwise patched Screen connect to make it less susceptible to abuse.
The speakers discussed a newly released bypass for a Defender patch and identified the researcher as Abdel Hammed Naseri.

Topics: [09:09] Reported Ruby Gems attack and AI-agent containment
[13:37] Limits on agent turns and air-gapped testing
[32:52] Project Lily, prompt review and model alignment
[43:24] LG TV data-collection claims and consent
[48:02] Passkey-themed phishing and Screen connect patch
[49:01] Defender patch bypass and security research