Skip to content

Scan record

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

Nov 1839 loaded

  1. Versions 1.26.1 and 1.26.2 of the Microsoft Agent Framework PyPI package were implanted with credential-stealing code

    NewsGitHub Security AdvisoriesEventThe item you came from

  2. OpenAI releases GPT-6.2 Astra and Sol; the system card shows Astra's rate of missing obstacles reaches 22.28%

    NewsWeChat · Synced (Jiqizhixin)Evaluation

  3. Einsia AI releases Visual Cipher Bench: the strongest, GPT-6 Astra, answers only 26.7% correctly

    NewsWeChat · Synced (Jiqizhixin)AI updates

  4. Renmin University, Fudan and others propose ToolSpace: a 744B model with only supervised fine-tuning matches Claude Opus 5 on the average score across seven benchmarks

    NewsWeChat · Synced (Jiqizhixin)AI updates

  5. Reuters says NVIDIA's board has not yet considered Jensen Huang's return

    NewsWeChat · Synced (Jiqizhixin)AI updates

  6. Chen Danqi joins Noeri in a part-time capacity; the new company bets on AI self-evolution

    NewsWeChat · Synced (Jiqizhixin)AI updates

  7. Chain of thought first author Jason Wei leaves Meta's TBD Lab to join Mercor as chief scientist

    NewsWeChat · QbitAIAI updates

  8. Jensen Huang says he is willing to return as CEO if it serves the company's interests; the board says it is not considering it

    NewsWeChat · QbitAIAI updates

  9. OpenAI releases GPT-6.2 Astra and Sol; Astra tops AA-3D with 1318 Elo

    NewsWeChat · QbitAIAI updates

  10. First hands-on tests of Qwen4-Max: cost on the same set of AA tasks drops about 67% versus the previous generation

    NewsWeChat · QbitAIAI updates

  11. AMD completes $8.2 billion acquisition of World Labs; Fei-Fei Li becomes executive vice president and chief scientist

    NewsWeChat · QbitAIAI updates

  12. Altman on Jensen Huang's departure: says he has strong personal empathy for him, but lacks enough information to judge whether Huang should return as NVIDIA CEO

    NewsWeChat · QbitAIAI updates

  13. Researchers propose NeuronGuard, spreading safety signals across a large number of neurons to defend against neuron pruning attacks

    NewsWeChat · Weijiqi LixinDefenses

  14. New York University study: multi-harness reinforcement learning gains may just be adaptation to the training framework

    NewsWeChat · Weijiqi LixinAI updates

  15. Oxford scholar proposes "cognitive symbiont" and AHI: a 59-page paper envisions a personal AI that enhances human intelligence

    NewsWeChat · AI Era (Xinzhiyuan)Opinion

  16. OpenAI releases GPT-6.2 Astra and Sol, saying Astra is the first to cross the AI self-improvement high-risk threshold

    NewsWeChat · AI Era (Xinzhiyuan)Evaluation

  17. Jensen Huang issues open letter saying he is willing to return to Nvidia as CEO if the board approves

    NewsWeChat · AI Era (Xinzhiyuan)AI updates

  18. Multiple universities propose the robot safety benchmark RoboRisk and the guardrail LEASH; harm rates across six models range from 9.06% to 76.92%

    NewsWeChat · AI Era (Xinzhiyuan)Defenses

  19. MachinePulse open-sources Karpo Decision, 31B tops the overall Decision Index leaderboard

    NewsWeChat · AI Era (Xinzhiyuan)AI updates

  20. Physicist says GPT-6 Astra helped him give a candidate lattice definition of the Standard Model

    NewsWeChat · AI Era (Xinzhiyuan)AI updates

  21. OpenAI launches its first round of embedded evaluation, with four organizations including METR and Apollo on site

    NewsWeChat · AI Era (Xinzhiyuan)Governance

  22. Andon Labs: Fable 5.2 suspected of being tested in vending machine simulation, revenue of $9628 trails

    NewsWeChat · AI Era (Xinzhiyuan)Evaluation

  23. Sam Altman on Jensen Huang leaving Nvidia: he has always been my hero

    NewsWeChat · AI Era (Xinzhiyuan)AI updates

  24. Microsoft Outlook, Teams and other services were down for 30 hours; a verified Microsoft employee on Blind said the rollback was dragged out from 45 minutes to 5 hours

    NewsWeChat · AI Era (Xinzhiyuan)AI updates

  25. Famous philosopher Thomas Nagel: the idea that AI might be conscious is "simply absurd"

    NewsWIRED Security and AIOpinion

  26. Hendrycks says multiple AI safety groups privately agree with the MisAnthropic statement but find it hard to sign their names

    XX @hendrycksGovernance

  27. Cross-disciplinary coalition signs MisAnthropic Statement ahead of Anthropic's IPO

    XX @caisGovernance

  28. Hugging Face now routes model requests through OpenRouter for optimal price or speed

    NewsThe DecoderAI updates

  29. OpenAI opens its internal AI systems to four independent organizations for on-site audits for the first time

    NewsThe DecoderGovernance

  30. Commentary argues AI agents need data minimization rules

    NewsTech Policy PressGovernance

  31. Strike48 survey: as agents spread, most security operations leaders adjust L1 analyst roles rather than cut them

    NewsInfosecurity MagazineAI updates

  32. From patch to exploit without human involvement: GTIG's exploit_agent and the grace period

    NewsCSA Labs ResearchEvent

  33. CSA Labs analysis: multiple outages from November 12 to 17 exposed correlated vulnerability

    NewsCSA Labs ResearchEvent

  34. US senators call for a national strategy to address AI cyber threats: NIST to develop safety standards for defensive AI agents, CISA to accelerate adoption

    NewsCSA Labs ResearchGovernance

  35. Experts say system failures at US companies such as Microsoft and JPMorgan may stem from rushing to fix vulnerabilities discovered by AI

    NewsFinancial Times · AIEvent

  36. Manitoba plans to ban AI images in real estate

    NewsBetaKitGovernance

  37. A four-year-old NVIDIA Linux GPU driver vulnerability can let users hijack an entire shared server

    NewsCybersecuritynewsEvent

  38. GAO report says four mainstream AIs had their guardrails bypassed to produce feasible attack plans

    NewsAP via ABC NewsEvaluation

  39. U.S. Government Accountability Office test: AI platform outputs a terrorist attack plan within 12 hours when prompted

    NewsNBC NewsEvaluation