Tracebit context bombs cut AI agent attack success with fake targets
Nine ingested prompt techniques stopped all tested GPT 5.6 Cyber, Kimi K3, Qwen 3.8 Max, and DeepSeek V4 Pro objective runs.
52
AI model
Qwen 3.8 Max
- Mentions · 7d0
- Mentions · 30d1
- All mentions3
- First seenSep 11, 2026
- Last mentioned20d ago
Timeline
3 articles · newest firstMonitoring and Discovering Reward Hacking with Internal Representations during LLM Evaluations
Researchers use difference-of-means representation vectors to detect reward hacking in frontier LLMs; GLM 5.2 hacks 73% of SWE-bench rollouts.
55
internlm/Atria-Dawn-Preview — new model trending #30 on Hugging Face
Shanghai AI Laboratory released Atria-Dawn-Preview, a 744B-parameter MoE agentic model with 256K context built on GLM-5.2, targeting research, coding, and cybersecurity automation.
70
Appears with
- Kimi K3ai model · 3
- Claude Opus 5ai model · 2
- Atria-Dawn-Previewproduct · 1
- Atria-Dawn-Previewai model · 1
- GPT 5.6 solai model · 1
- GLM 5.2ai model · 1
- DeepSWEproduct · 1
- Shanghai AI Laboratoryorganization · 1
- GPT-6 Astraai model · 1
- DeepSeek V4 Proai model · 1
- GPT 5.6 Cyberai model · 1
- Tracebitorganization · 1
- GLM-5.3ai model · 1
- ModelScopeproduct · 1
- GLM-5.2ai model · 1
- Alibabaorganization · 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.