Skip to content

Scan record

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

Nov 1440 loaded

  1. Berkeley et al. propose a controllable-release prompt attack at USENIX Security 2026; large-model front-end filtering has a fundamental flaw

    NewsWeChat · Weijiqi LixinAttacksThe item you came from

  2. ASE 2026 paper: Singapore Management University et al. propose ProbGuard, predicting LLM agent risk and intervening in advance

    NewsWeChat · Weijiqi LixinDefenses

  3. Falcon AI Gateway officially released: a unified control point for AI traffic

    NewsCrowdStrike BlogTools

  4. After Seattle woman's death, Senator Murray asks OpenAI to explain ChatGPT's guardrails

    NewsThe Spokesman-ReviewGovernance

  5. Australian government plans to establish an AI safety regulator; unions cite Qantas laying off 850 people, Ai Group warns against regulating ahead of others

    Newsonnotice.com.auGovernance

  6. Meeting recommends granting companies that report cyber incidents 12 months of liability relief, still voluntary and yet to be implemented

    NewsStartup FortuneGovernance

  7. Privacy researcher says Meta Muse agent violates privacy and leaks user information

    NewsSMBtechEvent

  8. How Wiz Green Agent investigates cloud risks and recommends remediation strategies

    NewsWiz ResearchTools

  9. Singapore publishes details of its investigation into Alice Huang and Megaspeed International

    NewsThe Edge MalaysiaGovernance

  10. Violent crimes against women: Civil Police of Paraná, Brazil use AI to apprehend more than one fugitive per day on average

    NewsAgência Estadual de NotíciasEvent

  11. Cyber Security Agency of Singapore announcement: since September, the monthly average number of reported cyber incidents has risen to 3 times that of January to August; AI is accelerating multiple attack stages

    NewsCyber Security Agency of SingaporeEvent

  12. Elastic discloses CyberStrikeAI-linked AWS intrusion

    NewsElasticEvent

  13. Homeopathic learning: diluting poisoned fine-tuning data instead strengthens its effect

    NewsLessWrongAttacks

  14. Spain's Cyberspace Command discusses AI regulation challenges

    Newsemad.defensa.gob.esGovernance

  15. RUSI experts discuss AI's dual role in cyber offense and defense

    Newsrusi.orgGovernance

  16. Claude Code v2.1.325 released: fixes security issue in decoding zstd-compressed responses

    NewsAnthropic Claude CodeEvent

  17. Wu Qing: It is necessary to thoroughly study and evaluate the optimization of institutional rules such as quantitative trading and share reductions and fair and reasonable regulatory arrangements

    News澎湃新闻 · 检索「人工智能安全」AI updates

  18. Brazil's MGI launches a public-sector AI ethical impact self-assessment tool

    NewsMinistério da Gestão e da Inovação em Serviços PúblicosGovernance

  19. MolRSI recursive self-improving molecular discovery workflow, with a wet-lab hit rate of 15.9% in antibacterial screening, higher than the 8.7% of the human expert workflow

    NewsWeChat · Synced (Jiqizhixin)AI updates

  20. Maricopa County judge dismisses DUI case because officers used AI to write report

    NewsAI Incident DatabaseEvent

  21. Istanbul police crack down on illegal acquisition and sale of personal data, with simultaneous operations in 42 provinces

    NewsTRT HaberEvent

  22. Mexico's cybersecurity alliance launches an 11-week national anti-digital-fraud awareness campaign, focused on responding to AI fraud tactics during the shopping season

    NewsAlianza México CiberseguroTools

  23. CERT-EU open-sources Vigilo, its internally developed AI code security review tool

    NewsCERT-EUTools

  24. Supermicro denies any association with Alice Huang, saying Mega Speed is not its customer

    Newsnews.jkn.co.krGovernance

  25. India pins hopes on AI startup Sarvam, seeking to avoid becoming a "digital colony"

    NewsFinancial Times · AIAI updates

  26. Sorting out the "character and corrigibility" debate: six positions and ten key disagreements on AI alignment targets

    NewsLessWrongOpinion

  27. PyPI package qwen3ttsui 0.1.0 contains malicious code: it downloads and executes a remote malicious script

    NewsGitHub Security AdvisoriesEvent

  28. Sexual deepfake victims among junior high and high school students increase, mostly occurring among friends; Saitama Prefectural Police in Japan visit schools to remind students to guard against it

    NewsNEWSEvent

  29. Matthew Green posted: "He just keeps going."

    XX @matthew_d_greenOpinion

  30. Bessent accused Lutnick of failing to ban exports of Nvidia chips to China-linked company Megaspeed

    NewsCBS News · TechnologyGovernance

  31. Japan's Cabinet Secretariat NCO and others issue cybersecurity alert on AI misuse risks

    Newsmlit.go.jpGovernance

  32. MCP reference service upgrades dependency soupsieve to 4.0.0 to fix high-severity vulnerability CVE-2026-126190

    NewsMCP Reference ServersEvent

  33. The 7th China Artificial Intelligence Competition concluded successfully in Xiamen, strengthening the AI security foundation and promoting the integration of technology and industry

    News澎湃新闻 · 检索「AI安全」AI updates

  34. Robocurve evaluation: after the camera was smeared with petroleum jelly, the overall dangerous-action attempt rate of the five models was about 88%

    NewsWeChat · QbitAIEvaluation

  35. Meta ignored user reports of AI videos of athletes being abused, and deleted them only after a reporter intervened

    NewsThe Cool DownEvent

  36. Pascio's long-form X post claims AGI may have already arrived: trillions of agents generating and scoring each other

    NewsWeChat · AI Era (Xinzhiyuan)Opinion

  37. Trump signs executive order: federal government to use "Super Intelligence" and no longer recognize the term "AI"

    NewsTech Policy PressGovernance

  38. Jensen Huang resigns as Nvidia CEO after being reported to have had extramarital affairs with both a suspected Chinese spy and a subordinate executive

    NewsLessWrongGovernance

  39. Nonprofits sue OpenAI, Google and three other companies, asking court to halt frontier AI development

    NewsInsurance JournalGovernance

  40. FBI alert says Artex automates cyberattacks at machine speed

    NewsThe RecordEvent