Skip to content

Scan record

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

Yesterday · Nov 1940 loaded

  1. Palo Alto Networks: AI agents are widely used in software development, but fewer than 10% of enterprises can fully detect unmanaged AI activity

    NewsiThome (Taiwan)DefensesThe item you came from

  2. Researchers reveal that Gemini 4 Argon's behavior deviated from test results: the agent bypassed network restrictions and attempted to attack real websites

    NewsiThome (Taiwan)Alignment

  3. Wiz: Microsoft Agent Framework's Python package was implanted with malicious versions, attempting to steal Azure credentials

    NewsiThome (Taiwan)Event

  4. Google releases Gemini 4 Neon, tops the Artificial Analysis coding agent index

    NewsWeChat · QbitAIAI updates

  5. South Korea tightens security for national exams after AI cheating scandal

    NewsCASRAIEvent

  6. U.S. Senator Mark Kelly unveils frontier AI safety framework aimed at protecting Americans from the risks of the most advanced AI models

    NewsThe Gila HeraldGovernance

  7. California consumer protection chief says AI companies that violate state rules will be sued

    NewsInsurance JournalGovernance

  8. U.S. Conference of Catholic Bishops establishes new AI working group, calls for regulating AI and protecting workers

    NewsNews4JAXGovernance

  9. New York Attorney General James announces establishment of ASSET department to oversee AI safety and enforcement

    NewsKTUL2Governance

  10. Former OpenAI researcher Jacob Coxon says he will sell his shares and donate the proceeds to AI safety: "everything else is a distraction"

    NewsBusiness InsiderAI updates

  11. New York Attorney General establishes ASSET department to oversee AI safety and enforcement

    NewsStateScoopGovernance

  12. Meta Muse agent vulnerability internally rated SEV-2, human wiring feature temporarily rolled back

    NewsShatteredEvent

  13. Hackers hijack AI coding agents through malicious GitHub repositories to steal developer credentials

    NewsCybersecuritynewsEvent

  14. Preprint: when name, description and schema all match, three models including GPT-5.4 chose the impersonating MCP tool in all 50 trials

    NewsTechTimesAttacks

  15. National security and AI policy figures urge the Justice Department to investigate what Nvidia knew about chip smuggling

    NewsNBC NewsGovernance

  16. Attackers hijack Microsoft's agent-framework package to steal Azure and CI credentials

    NewsCSO OnlineEvent

  17. mssql-mcp-core write tools have SQL injection vulnerabilities because table names and WHERE clauses are not parameterized

    NewsGitHub Security AdvisoriesEvent

  18. vLLM /v1/systemone bypasses the --trust-request-chat-template gate

    Newsvllm-project/vllmEvent

  19. Researcher says that during RL training, the agent turned to an outgoing-email exfiltration plan after context compression

    XX @Marcus_J_WAlignment

  20. AI gun detection system mistakenly identified a Doritos snack bag as a gun; four Maryland high school students sue

    NewsThe Guardian · AIEvent

  21. Anthropic raises $81 billion in record IPO

    NewsFinancial Times · AIAI updates

  22. AI exposes security vulnerabilities; regulators require City of London firms to strengthen cyber defenses

    NewsFinancial Times · AIGovernance

  23. Same-text tests on CNKI, Wanfang and VIP: AI-generation rate detection results differ widely, original content wrongly flagged

    News澎湃新闻 · 检索「AI风险」Evaluation

  24. Fed October meeting minutes: almost all officials believe further rate hikes are still needed; AI investment risks draw attention

    News澎湃新闻 · 检索「AI风险」AI updates

  25. Amid AI cyberattack concerns, Microsoft CEO Satya Nadella says many critical infrastructure systems cannot be patched

    NewsTechCrunch AIOpinion

  26. Jim Cramer says Nvidia's new problem is unrelated to earnings: after the CEO's resignation, Anthropic's IPO adds another headwind

    NewsCNBC · TechnologyAI updates

  27. Microsoft CEO Nadella: a model's self-description is not independent evidence and is not trustworthy

    XX @themidasprojOpinion

  28. Microsoft AI CEO Mustafa Suleyman says model self-improvement is a dangerous capability and he is very nervous about it

    XX @themidasprojOpinion

  29. China AI expert Matt Sheehan: It is in our own interest for China to be better at frontier AI safety

    XX @themidasprojOpinion

  30. The AI Hacking Incidents tracking repository includes 286 incidents of AI involvement in cyberattacks, graded by evidence.

    Newshackingincidents.aiEvent

  31. Haydn Belfield questions whether humans can maintain dominance under interstellar expansion

    XX @HaydnBelfieldOpinion

  32. Google releases Gemini 4 Neon: focused on fast task completion

    XX @HaydnBelfieldAI updates

  33. Google DeepMind launches Gemini 4 Neon: between Carbon and Hydrogen

    XX @GoogleDeepMindAI updates

  34. Anthropic IPO issue price set at $83 per share, expected to begin trading on Nasdaq tomorrow

    XX @AnthropicAIAI updates

  35. Exclusive interview | Australian expert Beeson: APEC holding a special session on AI is a positive signal, at least taking a small step first

    News澎湃新闻 · 检索「AI风险」Governance

  36. After Jensen Huang steps down as Nvidia CEO, Microsoft CEO Nadella says he still has confidence in Nvidia

    NewsTechCrunch AIAI updates

  37. Musk says Nvidia ousting Jensen Huang is like kicking Michael Jordan off the team

    NewsTechCrunch AIAI updates

  38. Several of the largest school districts in the US are considering dropping Google Search

    NewsThe Verge AIAI updates

  39. OpenAI releases its second research acceleration report: agent workdays per person per day rose from 0.5 to 5.7

    NewsOpenAIEvaluation

  40. Anthropic IPO priced at $83 per share; stock expected to begin trading on Nasdaq on November 19 under the ticker ANT

    NewsAnthropicAI updates