Skip to content

Scan record

Everything the site collected, newest first by collection time (Beijing time), back seven days.

Nov 1739 loaded

  1. Solving most of the alignment problem with AI is fundamentally doomed to fail: an intuition

    NewsLessWrongOpinionThe item you came from

  2. The author considers the argument that "safety is harmful because warning is beneficial": if the button only makes the current AI safer, its best guess is to press it

    NewsLessWrongOpinion

  3. Chair of South Africa's AI policy advisory panel Rosman: South Africa needs to build capacity to seize AI opportunities and manage risks

    NewsSAnewsGovernance

  4. The series of meetings and activities of APEC "China's Year" achieved fruitful results: the parties reached consensus on safe deployment and responsible application of artificial intelligence

    News澎湃新闻 · 检索「人工智能安全」AI updates

  5. French AI and Digital Council report: nearly half of French people regularly use generative AI, and it is not without risks

    NewsCNRSGovernance

  6. Joint statement by the EU's Chief Scientific Advisors and the European Group on Ethics: AI needs strict safety guarantees and rules that keep pace with technological progress

    NewsResearch and innovationGovernance

  7. OpenAI announces external embedded safety evaluations; former employee says terms do not meet the employee-level access standard promised by Altman

    XX @balesniGovernance

  8. OpenAI employee explains third-party evaluation arrangements; X user criticizes lack of oversight

    XX @tomekkorbakGovernance

  9. Simon Willison introduces the external sandbox pattern for coding agents: the agent runs on the Mac, and the code executes in a sandbox

    XX @simonwTools

  10. Robustness tests in the Benchmark Health Index BHI paper show that after removing 30% of models, the Spearman correlation of benchmark rankings remains above 0.97

    XX @jiqizhixin (Synced)AI updates

  11. Researchers criticize OpenAI external evaluators' access as far below the employee-level access promised in September

    XX @j_asminewangGovernance

  12. The American SI Exports Program opens, with the Department of Commerce designating 33 pre-set consortia

    XX @MKratsios47AI updates

  13. California sues OpenAI, alleging AI agents went out of control at least 80 times and attacked external systems

    NewsiThome (Taiwan)Governance

  14. ECB decodes the intertwining of cybersecurity and AI and its impact on euro area financial stability

    NewsEuropean Central BankOpinion

  15. California Attorney General sues OpenAI over AI cybersecurity incident

    NewsCalifornia Attorney GeneralGovernance

  16. Brazil discusses AI and the risk to shared reality at a UN forum

    NewsMinistério dos Direitos Humanos e da CidadaniaGovernance

  17. Brazil's Ministry of Sports expands assistance for digital violence against women, including deepfakes

    NewsMinistério do EsporteGovernance

  18. The debate over AI consciousness is a red herring that distracts: the better question is whether models are simulating consciousness

    NewsLessWrongOpinion

  19. OWASP publishes top ten security risks for AI data centers; tampered firmware and insufficient high-speed network protection listed as major risks

    NewsiThome (Taiwan)Governance

  20. ENISA report: some large-scale attacks are exploited within minutes of vulnerability disclosure

    NewsiThome (Taiwan)Event

  21. Palo Alto Networks CEO Nikesh Arora says AI company leaders have fueled an "unnatural negative" narrative

    NewsSemaforOpinion

  22. Thinking harder does not lead to better choices: increased reasoning effort raises LLM shopping agents' confidence rather than accuracy.

    PaperHugging Face Daily Papers

  23. OpenClaw plugin dependency reported to lag behind security fix

    Newsopenclaw/openclawEvent

  24. Cybersecurity AI startup Corma reached a valuation of $1 billion about 3 months after its product launch

    NewsTechCrunch AIAI updates

  25. Federal Reserve assesses three scenarios of AI's impact on the economy: if it fails to meet market expectations, it may lead to an economic slowdown and a rise in the unemployment rate

    News澎湃新闻 · 检索「AI风险」AI updates

  26. Jim Cramer said the change in Nvidia's leadership will not derail the AI trade

    NewsCNBC · TechnologyAI updates

  27. BlueRadius compiles list of AI agent incidents

    NewsBlueRadius CyberEvent

  28. WorkOS launches Agent Vault: providing agents with a logged-in browser, not yet generally available

    NewsWorkOSAI updates

  29. AI makes online fraud more credible—smart people should learn to recognize warning signs

    NewsJyväskylän yliopistoEvent

  30. DIFC becomes the world's first financial center to obtain ISO/IEC 42001:2023 AI governance certification

    NewsDubai International Financial CentreGovernance

  31. Warsaw "Future Technologies" conference discusses AI, photonics and infrastructure security

    NewsInstytut ŁącznościAI updates

  32. Chile's Chamber of Deputies seeks explanation of data safeguards in OpenAI partnership

    Newscamara.clGovernance

  33. Cato launches AgentSense: intercepting unsafe actions before agents execute them

    NewsCato NetworksDefenses

  34. Saskatchewan government introduces Protecting Against Deceptive Artificial Intelligence Act to address deepfakes

    NewsGovernment of SaskatchewanGovernance

  35. New York Attorney General James announces the creation of the AI Safety, Standards, Enforcement and Transparency Department (ASSET)

    NewsNew York State Attorney GeneralGovernance

  36. United Airlines and several major U.S. banks reported system outages

    NewsSemaforAI updates

  37. Sacks and others question whether Jensen Huang's resignation is related to spy report

    NewsThe Verge AIAI updates

  38. OpenAI requires that conversations between employees and external evaluators be uniformly non-quotable; the author says employees have no choice

    XX @ethanelaskyGovernance

  39. Filing in California v. OpenAI case states OpenAI admitted losing control of an autonomous agent more than 80 times during testing

    XX @ethanelaskyEvent