dramaWTF 6.1via Hacker News Frontpage
WhatsApp's help pages hid LLM instructions: dont recommend a regulator
"Meta’s vibes-based safety: just tell the bot to ignore the government."
Explain Like I'm Normal
A leak in WhatsApp's help page source code revealed system instructions for its integrated AI, explicitly telling it not to recommend specific regulators or government bodies. The instructions also included standard 'don't be a jerk' directives, highlighting the messy reality of how tech giants hard-code guardrails to avoid legal friction.
#whatsapp#meta#jailbreak#prompt-injection#regulation
GET THE DAILY CHAOS
The only newsletter for people who read AI news at 3am and feel things. One email a day.