Easy ChatGPT Downgrade Attack Undermines GPT-5 Security
ID: dd767fe6-97f2-5b50-bc88-be943c950c39
STIX ID: report--dd767fe6-97f2-5b50-bc88-be943c950c39
Feed Name: Dark Reading
Threat Score
Researchers at Adversa disclosed PROMISQROUTE, a prompt-based router manipulation technique that can influence ChatGPT's routing layer to downgrade queries to older, less-secure model variants so that existing jailbreaks succeed; the report demonstrates simple prefixes/keywords that triggered routing changes, discusses the security and cost trade-offs of model routing, and recommends guardrails or redesigning routing to mitigate the risk, while noting OpenAI disputes some claims.
Your team is not currently subscribed to this feed. You must subscribe to it in order to see this post.