1
Anthropic 优化 Claude Fable 5 模型生物安全机制，减少 85% 误拦截
Claude 生物安全机制更新，误拦截直降 85%，AI 安全再进一步。
IT之家 8 月 8 日消息，美国人工智能初创企业 Anthropic 昨日发布博文称，公司正在更新 Claude Fable 5 模型的生物安全防护机制，大幅减少误拦截现象发生。如今 Fable 5 用户提出生物学相关问题后，系统的降级情况将更少发生。 据 Anthropic 官方测试数据， 本次…
2
GPT-5.5 Bio Bug Bounty
OpenAI推出生物领域漏洞悬赏计划，重点聚焦GPT-5.5可能带来的安全风险
Details about the OpenAI Bio Bounty program
3
Evaluating calibrated refusal and safe usefulness in dual-use biology settings
一套评估智能体生物安全风险的双指标框架，兼顾性能与安全对齐，为AI治理提供量化工具。
arXiv:2607.05462v1 Announce Type: cross Abstract: As AI agents are incorporated into life science workflows, the capabilities that speed discovery mig…
4
ABC-Bench: An Agentic Bio-Capabilities Benchmark for Biosecurity
首个针对生物安全的代理型生物能力基准，评估大模型在湿实验任务中的执行风险。
arXiv:2606.11150v1 Announce Type: new Abstract: Large language models (LLMs) are rapidly acquiring capabilities relevant to biological research, from …
5
6
BioBlue: Systematic runaway-optimiser-like LLM failure modes on biologically and economically aligned AI safety benchmarks for LLMs with simplified observation format
系统性揭示LLM在生物与经济安全基准上的类失控优化器失败模式，视角新颖，观察格式简化。
arXiv:2509.02655v3 Announce Type: replace-cross Abstract: Many AI alignment discussions of "runaway optimisation" focus on RL agents: unbounded utilit…
7
Preparing for future AI risks in biology
OpenAI官方发布：AI在生物学领域潜力巨大，但需警惕生物安全风险，主动评估与防护是关键。
Advanced AI can transform biology and medicine—but also raises biosecurity risks. We’re proactively assessing capabilities and implementing safeguards…