Every company has its own version of ChatGPT now
Summary: A Cisco Talos newsletter writer tested Meta's open-source Llama 2 and found it refused prompts seeking phishing emails and gift-card scam scripts, repeatedly citing its terms of service and the harm of spam and phishing. The same prompts given to ChatGPT produced a customizable email template asking the recipient to download an attachment, with no pushback.
Why it matters: One widely used chatbot declined to help with a phishing lure while another produced a repeatable template on request, which means the guardrails on the AI tools your employees already have access to are not consistent.
Read the original at Cisco Talos ↗
- Topics
- Phishing & Social Engineering
- Related
- US disrupts North Korean campaign behind $400 million crypto theft Cyber Risks · Nov 14, 2026North Korea ‘s Crypto Theft Operations: DOJ Seizes $1.2 Billion and Disrupts Hacker Infrastructure Cyber Risks · Nov 13, 2026FBI Says North Korean Hackers Stole Over $400 Million in Crypto Using Deepfake Video Calls Cyber Risks · Nov 13, 2026