#ai safety
2 articles
Latest
Research
Streaming LLM Agents Shift From Transaction to Revision Model
New theory reframes agent execution as continuous dialogue where users can intervene mid-task, not transaction completed in isolation.
18 days ago8 min read
Security
Anthropic's Mythos Model Poses Unprecedented Cybersecurity Threats
Internal testing reveals the model introduces novel attack surfaces and defense evasion capabilities.
about 2 months ago3 min read