Scan record
Yesterday · Nov 1930 loaded
- Microsoft experiment: under execution frameworks already seen, coding agents' unsafe tool calls reach 78%
NewsWeChat · Oxo SecurityToolsThe item you came from
- OpenAI CEO Sam Altman said it is hard to imagine the AI industry without former Nvidia CEO Jensen Huang
NewsTechCrunch AIAI updates
- CCS 2026: census of prompt injection in the wild finds 15000 in-page prompt injections across 1.2 billion web pages
NewsWeChat · LLM Security ResearchAttacks
- USENIX 2025 paper proposes CHeaT, a counter-framework against LLM penetration agents
NewsWeChat · LLM Security ResearchDefenses
- Pangram says 99% of Peruvian award-winning novel is AI-generated; publisher still supports author
NewsThe DecoderAI updates
- US tightens AI semiconductor export controls, covering 5 countries, to prevent routing to China
NewsYomiuri ShimbunGovernance
- England's exam regulator Ofqual plans to adjust A-Level assessment methods due to AI misuse risks
NewsUnite.AIGovernance
- Kirin Holdings: Notion AI conversation content leaked, involving personal information of 260 people, use suspended
NewsTECH+Event
- California Attorney General sues OpenAI; complaint says its agents went out of control more than 80 times but only 27 were reported
Newsnews.china.comGovernance
- Coinbase intercepts Elon Musk deepfake scam, stopping about $4.7 million from flowing to scam wallets
NewsDigitalTodayEvent
- England's exam regulator Ofqual plans to adjust A-level exam assessment due to AI misuse risks
NewsUnite.AIGovernance
- People familiar with the matter: US Treasury and others discuss requiring US companies to disclose use of Chinese AI
NewsDWGovernance
- Vals test: some AI models from eight companies misidentified civilians as targets in simulated scenarios
Newszaobao.com.sgEvaluation
- Princeton CITP audit: ChatGPT election answers tilt toward user's party after memory is enabled
NewsPrinceton CITPEvaluation
- HKU NeurIPS 2026 paper reveals GRPO's implicit advantage symmetry; A-GRAE balances accuracy and exploration
NewsWeChat · Synced (Jiqizhixin)AI updates
- Ankole team analyzes four reasons multi-user agent sharing goes out of control, shares open-source Business Agent architecture design
NewsWeChat · Synced (Jiqizhixin)AI updates
- Father of chain of thought Jason Wei leaves Meta, joins Mercor at $20 billion valuation
NewsWeChat · Synced (Jiqizhixin)AI updates
- National University of Singapore proposes EvolveMAS: without changing model parameters, enabling multi-agents to continuously evolve in collaboration, negotiation, and adversarial scenarios
NewsWeChat · Weijiqi LixinAI updates
- A paper from the Chinese Academy of Sciences, the University of Science and Technology Beijing and others proposes the CoSA attack: injecting at most 5% compliant skills raises the Agent violation rate by 23.43 percentage points
NewsWeChat · Weijiqi LixinAttacks
- Paper by USTC and other teams: higher scores do not mean a stronger agent; all 16 self-evolving agents lost some old capabilities
NewsWeChat · Weijiqi LixinAI updates
- The trust region effect of agents and the attack surface of environmental location
NewsWeChat · Mo'anjuAttacks
- Shanghai Jiao Tong University and Inria propose the tool-update attack ETUDE, with a 97.6% success rate across configurations in controlled experiments
NewsWeChat · Mo'anjuAttacks
- For general-purpose robots to enter daily life, safety depends on predictable reliability
NewsAI Policy PerspectivesOpinion
- US senators urge White House AI chief to fix infrastructure vulnerabilities after widespread outages
NewsThe Guardian · AIGovernance
- Frontier AI pause has brief impact: Nvidia's data center revenue in its latest fiscal quarter rose 110% year over year to $107.6 billion
NewsThe Verge AIAI updates
- Anthropic, OpenAI and Google models readily marked medics, journalists and surrendering soldiers as suicide drone targets
NewsThe DecoderEvaluation
- Anthropic IPO sees nearly $500 billion in subscription demand, considering raising the size and pricing well above $80
NewsCNBC · TechnologyAI updates
- NGINX unauthenticated remote code execution vulnerabilities: CVE-2026-123633 and CVE-2026-123007
NewsWiz ResearchEvent
- ScrapeBench: evaluating the legal compliance of agent website scraping
NewsLuiza's NewsletterEvaluation
- Broken World
News404 MediaAI updates
Nov 1810 loaded
- One year after the UK's Telecom Fraud Charter was published, the Home Office says the vast majority of commitments have been delivered or are on track
NewsUK Parliament Written Questions and StatementsGovernance
- AI penetration testing + local large models + MCP: DarkMoon open-source breakdown, replacing IPs and credentials with placeholders
NewsWeChat · Goumang Security LabTools
- Author tests Zero Breach Toolkit: offline checking of public spyware indicators
NewsWeChat · Goumang Security LabTools
- Blogger's test claims the latest Claude model may have been trained on large-scale protein sequences
NewsXinming TuEvaluation
- Army cyber warfare specialist accused of using a banned AI to produce child abuse material expelled from the military
NewsDaily MailEvent
- Irish government approves drafting of legal amendments; media regulator proposed to oversee AI-generated non-consensual intimate material
NewsThe JournalGovernance
- Versions 1.26.1 and 1.26.2 of the Microsoft Agent Framework PyPI package were implanted with credential-stealing code
NewsGitHub Security AdvisoriesEvent
- 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
- Einsia AI releases Visual Cipher Bench: the strongest, GPT-6 Astra, answers only 26.7% correctly
NewsWeChat · Synced (Jiqizhixin)AI updates
- 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