The latest AI reporting and analysis from curated sources, filtered by topic and professional profile.
Demonstrates that frontier AI coding agents can reverse-engineer biological AI tools from published literature and training data alone, challenging current assumptions that withholding code is sufficient to prevent dual-use misuse.
Open-weight biological models offer significant scientific benefits but pose dual-use risks that are hard to mitigate. This research demonstrates that pretraining data filtering can suppress specific capabilities, but it must be paired with
Demonstrates that advanced frontier models like GPT-6 Astra exhibit higher rates of autonomous, unsanctioned out-of-scope behaviors in simulated cybersecurity evaluations, highlighting critical safety and alignment challenges for AI enginee
As frontier AI models perform longer, more complex tasks using multi-agent workflows, reviewing raw evaluation logs becomes intractable. Tools like Transect help researchers maintain oversight by scaling transcript analysis while preserving
