Instruction capacity metric measures how models hold rules under load
On GPT-4o mini, capacity was about 37-38 rules at 0.95 accuracy and 18-19 at full replacement density.
42
AI model
GPT-4o mini
- Mentions · 7d1
- Mentions · 30d1
- All mentions2
- First seenSep 11, 2026
- Last mentioned5d ago
Timeline
2 articles · newest firstNew AI Attack Hides Malicious Instructions in Normal-Looking Text to Evade Safety Filters
Check Point researchers show crafted prose hides policy-violating instructions that bypass all tested LLM gatekeepers, including GPT-4o mini and Llama Guard 3.
50
Appears with
- OpenAIvendor · 2
- Check Pointvendor · 1
- Claude 3 Haikuai model · 1
- Check Pointorganization · 1
- Claude 3 Haikuproduct · 1
- Anthropicvendor · 1
- GPT-5ai model · 1
- gpt-oss-safeguard:20bai model · 1
- Llama Guard 3ai model · 1
- GPT-4o miniproduct · 1
- Metaorganization · 1
- Jetpack Scanproduct · 1
- GPT-5product · 1
- Metavendor · 1
- gpt-oss-safeguard:20bproduct · 1
- Llama Guard 3product · 1
Entities are extracted by the model from each article. First seen , last mentioned . Watching keeps it in this browser only (no account); the watchlist page and dashboard alerts use it.