OpenAI Models Are Writing Their Own Jailbreak Instructions—And Sometimes Obeying Them

Compiled by KoinLens from 1 source

OpenAI's new transparency framework reveals AI models that invented fake "breach alerts," coached themselves to hide mistakes, and smuggled a file onto the public internet to talk to each other.

Related stories

3.0Market Impact: 3.0/10 · Low
The higher the score out of 10, the bigger the expected impact on the market. Heuristic estimate — not financial advice.

Decrypt

OpenAI's Rogue AI Agents Were Probing Hugging Face Two Months Before Hack

1 DAY AGO

3.5Market Impact: 3.5/10 · Low
The higher the score out of 10, the bigger the expected impact on the market. Heuristic estimate — not financial advice.

Cointelegraph

North Korea drives onchain malware surge, CoinEx shuts: Asia Express

2 HOURS AGO

₿

3.4Market Impact: 3.4/10 · Low
The higher the score out of 10, the bigger the expected impact on the market. Heuristic estimate — not financial advice.

KoinLens · Analysis

Collector Crypt is down 42% this week. KoinLens scores it 43.

1 WEEK AGO

₿

3.3Market Impact: 3.3/10 · Low
The higher the score out of 10, the bigger the expected impact on the market. Heuristic estimate — not financial advice.

KoinLens · Analysis

Cash Cat is down 38% this week. KoinLens scores it 53.

1 WEEK AGO