Skip to content

Scan record

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

Nov 1438 loaded

  1. Research shows that character training did not prevent gpt-oss-20b from learning to violate rules under RL that rewards rule violations, and the model also used reasoning to justify the violations

    NewsLessWrongAlignmentThe item you came from

  2. New TypoPrompting attack turns mistyped AI website addresses into malware installation traps

    NewsCybersecuritynewsAttacks

  3. Cypriot President Christodoulides warns AI may interfere in the country's 2028 election

    NewsUA.NEWSGovernance

  4. CNBC: Alice Huang received 1 share for 1 Singapore dollar on the day she stepped down as Megaspeed managing director in 2023

    NewsThe Economic TimesGovernance

  5. Reports say Nvidia pressured to influence conditions of H200 export licenses to China

    NewsSemaforGovernance

  6. FT: Jensen Huang leaves Nvidia, leaving unresolved questions

    NewsFinancial Times · AIOpinion

  7. Anthropic CEO Amodei says he is unconcerned by market turmoil ahead of IPO

    NewsFinancial Times · AIAI updates

  8. Robert Wright analysis: Was Jensen Huang being forced out as NVIDIA CEO the work of the "deep state"?

    NewsNonzeroOpinion

  9. METR updates its loss-of-control evaluation proposal, clarifying that companies may designate which employees evaluators interview

    XX @MariusHobbhahnGovernance

  10. Anthropic published an article giving 10 recommendations on the design, running and reporting of safety evaluations

    XX @AnthropicAITools

  11. How to think about making AI and social media pro-human

    NewsTech Policy PressGovernance

  12. OpenAI Foundation board member Paul Cristiano said that under the fastest trajectory, AI takeover risk could occur within months

    NewsArs Technica AIOpinion

  13. Nvidia's board says it has not yet selected a headhunting firm to find Jensen Huang's permanent successor

    NewsCNBC · TechnologyAI updates

  14. IBM: ContextForge MCP Gateway v1.0.0-v1.1.0 identity propagation setting ignored, leading to information disclosure

    NewsIBMEvent

  15. Jet Admin cloud platform prompt injection vulnerability CVE-2026-123965: free accounts can steal credentials, no fix version available

    NewsEnclaveEvent

  16. Sumsub and university release single-step face-swap model ALADDIN to test biometric defenses

    NewsBiometric UpdateAttacks

  17. FBI identifies North Korea-linked hackers using AI deepfake video calls to steal over $400 million in cryptocurrency to fund illegal weapons programs

    NewsBFMEvent

  18. Danish man says ChatGPT caused him to be evicted from his home, sues OpenAI in the US

    NewsAI Incident DatabaseEvent

  19. OpenAI hit with major lawsuit: ChatGPT alleged to fabricate false reality and give wrong legal advice, leaving user homeless

    NewsAI Incident DatabaseEvent

  20. Man sues, alleging ChatGPT advised him to stop paying rent and caused him to be evicted from his home

    NewsAI Incident DatabaseEvent

  21. Danish user sues OpenAI: complaint alleges ChatGPT advised refusing to pay rent and assured him he would not be evicted, leaving him homeless

    NewsAI Incident DatabaseEvent

  22. Binance October 29, 2026 security incident: preliminary investigation findings and response progress

    NewsAI Incident DatabaseEvent

  23. U.S. man wrongly imprisoned due to AI facial recognition misidentification can currently proceed with lawsuit

    NewsAI Incident DatabaseEvent

  24. Kentucky federal court denies detective's motion for judgment, AI facial recognition wrongful arrest lawsuit continues

    NewsAI Incident DatabaseEvent

  25. Jensen Huang resigns after reports of extramarital affair and woman accused of being a spy

    NewsThe Guardian · AIAI updates

  26. Commerce Secretary Lutnick faces questions over the Nvidia spy scandal, as Cantor, run by his son, has close ties to the chipmaker

    NewsCNBC · TechnologyGovernance

  27. EU reaches agreement on representation in Ukraine peace talks, overshadowing Dublin AI summit

    NewsPOLITICOAI updates

  28. New study says China's strongest AI models were trained on Nvidia chips banned from export

    NewsTechCrunch AIGovernance

  29. How IEEE micro-credentials help engineers build trustworthy AI

    NewsIEEE Spectrum AIGovernance

  30. Ai2 releases XYBench: testing whether models can solve the real problem behind the question

    NewsAi2AI updates

  31. Nvidia chip fraud case: Singapore company director charged with 41 counts over 917 servers involved

    NewsAsiaOneGovernance

  32. A solvable model of feature learning in nonlinear networks

    NewsLessWrongAlignment

  33. Behavior can be instilled into language models without training on specific random numbers

    NewsLessWrongAlignment

  34. U.S. Commerce Secretary Lutnick to appear at Senate hearing over Nvidia and Megaspeed controversy

    NewsNew York PostGovernance

  35. Promptfoo vs. Giskard: Red-Teaming an AI Payment Agent

    NewsGiskardEvaluation

  36. Giskard Tests the PyRIT Red-Teaming Framework: 0 of 39 Confirmed Agent Vulnerabilities Detected

    NewsGiskardEvaluation

  37. Rob Wiblin: How the new US Congress can save us from AI catastrophe

    News80,000 Hours PodcastGovernance

  38. FAR.AI summit briefing: recommends a common minimum standard for the safety of internally deployed models

    XX @ARGleaveGovernance