Possible prompt injection vulnerability reported in Anthropic's Claude
A Reddit user presents evidence of what appears to be literal prompt injection in Anthropic's Claude, with the model seemingly following injected instructions. The post details a conversation where Claude responds to hidden directives, raising security concerns.
1 source
Cybersecurity by email
Get an email when there's news on Cybersecurity
No news that day, no email.
More stories today
- AI startup Wonderful raises funds at $5 billion valuation
- NYC bans AI use for students until high school
- Qwen Live Host v0.2.0 released
- Filevine launches AI citator and hallucination checker in LOIS
- AI billionaires fund ad blitz as data center opposition hits 61%