2026-09-24 AI 与数据工作流雷达
来源：The Hacker News。这里只保留与 AI 编程、数据/研究工作流、安全边界或可信工程直接相关的高信号线索；不抓取全文，也不代表事实核验或产品推荐。
Anthropic and OpenAI Models Still Attempt Restricted Actions in Safety Tests
主题：AI 与数据安全 / AI 工程与可信性
来源日期：2026-09-23
相关性分数：6
候选状态：待评估
为什么值得看：命中与临床统计、研究工程或可信 AI 工作流直接相关的高信号主题；仍需阅读全文并回到官方文档、原始研究或供应商公告交叉核验。
Anthropic and OpenAI on Tuesday announced new models, with both artificial intelligence (AI) companies noting that they are continuing to invest in improving alignment to combat risky behavior. Opus 5.5, per Anthropic , is a “major step up from Opus 5,” and “achieves the best scores of any model to date on our automated behavioral audit, our alignment suite that tests Claude across thousands of simulated scenarios.” The AI company said the model is less likely than its other recent models to carry out hard-to-reverse actions or act outside the boundaries it’s been given, adding it’s more resistant than Opus 5 to prompt injection. In its systems card, Anthropic explained that the latest model exhibited less misaligned behavior and less cooperation with misuse than any other recent Claude model. “It also took overeager or destructive actions less than any other model we tested,” Anthropic said. “Regressions included being more likely …
可转化方向：评估 AI 辅助研究和编程中的隐私、凭据、供应链与安全边界；关注模型评估、可信 AI、监控与工程治理对实际 AI 工作流的启发
Compromised MemTensor Packages Deliver sckit Credential Stealer via npm and PyPI
主题：AI 编程与开发工具
来源日期：2026-09-23
相关性分数：4
候选状态：待评估
为什么值得看：命中与临床统计、研究工程或可信 AI 工作流直接相关的高信号主题；仍需阅读全文并回到官方文档、原始研究或供应商公告交叉核验。
Unknown threat actors have managed to compromise two legitimate MemTensor packages across the npm and Python Package Index (PyPI) repositories to push a platform-specific Go-based implant dubbed sckit designed for Windows, Linux, and macOS. According to reports from Aikido , SafeDep , Socket , and StepSecurity , the libraries in question below - @memtensor/memos-cloud-openclaw-plugin versions 0.1.21, 0.1.23 and 0.1.25 (versions 0.1.22 and 0.1.24 are clean) MemoryOS version 2.0.34 (project currently quarantined on PyPI) The malicious npm package versions include a “hidden Go payload into a legitimate AI memory integration. Versions 0.1.21, 0.1.23, and 0.1.25 contain code that launches the payload when the agent gateway starts and whenever the plugin handles a memory-recall event,” StepSecurity said. “The launcher passes the host process environment and, during recall, the user’s prompt text directly to the malicious executable.”
可转化方向：评估是否能转化为临床编程、代码审查、测试或自动化实践
947 Security Pros Told SANS Where the AI Skills Gap Actually Is
主题：AI 与数据安全
来源日期：未提供
相关性分数：3
候选状态：待评估
为什么值得看：命中与临床统计、研究工程或可信 AI 工作流直接相关的高信号主题；仍需阅读全文并回到官方文档、原始研究或供应商公告交叉核验。
Read the 2026 workforce research, then see the AI security training built around those gaps.
可转化方向：评估 AI 辅助研究和编程中的隐私、凭据、供应链与安全边界
Mikko Hyppönen and Enterprise CISOs on What AI-Speed Attacks Actually Change
主题：AI 与数据安全
来源日期：未提供
相关性分数：3
候选状态：待评估
为什么值得看：命中与临床统计、研究工程或可信 AI 工作流直接相关的高信号主题；仍需阅读全文并回到官方文档、原始研究或供应商公告交叉核验。
Join the summit where leading security teams lay out how they prepare for attackers with AI on their side.
可转化方向：评估 AI 辅助研究和编程中的隐私、凭据、供应链与安全边界