Hazel welcomes back Ryan Fetterman from the SURGe team to explore his new research on how large language models (LLMs) can assist those who work in security operations centers to identify malicious PowerShell scripts. From teaching LLMs through examples, to using retrieval-augmented generation and fine-tuning specialized models, Ryan walks us through three distinct approaches, with surprising performance gains. For the full research, head to https://www.splunk.com/en_us/blog/security/guiding-llms-with-security-context.html

Chetan Raghuprasad joins Hazel to discuss his threat hunting research into fake AI tool installers, which criminals are using to distribute ransomware, RATS, stealers and other destructive malware. He discusses the attack chain of three different campaigns, including one which even tries to justify its ransom as "humanitarian aid."

For the full research, read Chetan's blog at https://blog.talosintelligence.com/fake-ai-tool-installers/

Edmund Brumaghin joins Hazel to discuss how threat actors (including state sponsored attackers), are increasingly compartmentalizing their attacks i.e they're bringing in specialist skillsets from other groups to handle different aspects of the attack chain. Edmund discusses why this is happening, and the challenges this poses for defenders when it comes to attribution and reporting. He then discusses several solutions which seek to evolve traditional threat modelling, and help provide clarity to defenders.

More details can be found in this blog https://blog.talosintelligence.com/compartmentalized-threat-modeling/

If you're interested in our other blog on initial access groups, that can be found at https://blog.talosintelligence.com/redefining-initial-access-brokers/