Skip to content
Live
#Safety & alignment
Nov 14Sat
  1. ITHome · AI

    The White House's 'Superintelligence Working Group' calls for relaxation of accountability policies, encouraging AI companies to report cybersecurity incidents.

    The White House's 'Superintelligence Working Group' suggests that federal agencies relax accountability policies for AI cybersecurity incidents, limiting companies' regulatory responsibilities for security incidents in the next 12 months. The working group believes that current penalty standards may hinder companies from sharing security information, therefore encouraging AI companies to proactively report major security incidents. Previously, the working group had warned developers that they must report security incidents or face penalties.

  2. Interconnects65

    Huang Renxun's Departure and Its Impact on the Open Source AI Ecosystem

    Huang Renxun's departure has brought significant setbacks to the development of the US open-source model ecosystem. His support for open-source models was to avoid customer self-research of chips, which would lead to a decline in NVIDIA's profits, but supporting the open-source ecosystem is far more complex than technical issues. The economy of open-source models is still in its infancy and needs to break through in establishing use cases, revenue models, and continuously training next-generation models. NVIDIA's primary goal in the short term is to stabilize operations, deliver customer products, and prevent collapse, while the long-term value of open-source models has not yet been realized due to a shortage of computing resources.

  3. ITHome · AI65

    Uber hit by hack, driver license info leaked in multiple US states

    Ride-hailing platform Uber suffers data breach, thousands of drivers' license information stolen, involving Massachusetts, South Carolina, Rhode Island, and Illinois in the US. Hackers infiltrated third-party service provider Fireworks AI and stole employee credentials to access Uber's system and obtain data. This is Uber's second security incident in 2026, after Uber Freight was targeted by ransomware group Helix in August.

  4. ITHome · AI65

    Cursor disclosed the results of the security incident investigation, with the malicious Claude Code extension affecting up to 5000 users.

    Cursor announces investigation results of a security incident involving malicious Anthropic Claude Code extension, affecting up to 5000 users, mainly leaking link tracking data of users who did not use the privacy mode (less than 0.05%); attackers used Fireworks leaked keys to infiltrate some SaaS tools and planted malicious extensions for about 11 hours, Cursor has notified affected users and strengthened security protections.

  5. ITHome · AI65

    New Zealand's largest retailer The Warehouse accuses the use of AI to monitor employees' toilet time, sparking protests

    The Warehouse Group, New Zealand's largest retailer, has seen its employees protest the company's use of AI to monitor toilet time, forcing some employees to urinate and defecate in the delivery center corridors. This week, over 300 employees held protests, concerned about privacy violations from AI monitoring and potential layoffs due to the application of new technologies. Employees believe the company prioritizes robots over employees and worry about their job prospects.

  10. Where's Your Ed At58

    Premium: Unveiling the Transnational Network Suspected of Smuggling NVIDIA GPUs

    The Culper Research report accuses NVIDIA of channeling over 20% (approximately $43.8 billion) of its 2026 fiscal year revenue from China through a cross-border smuggling network, involving the personal relationships of executive Jensen Huang; Culper has gone short on NVIDIA due to this report, facing significant risks; mainstream media like The Wall Street Journal have reported inadequately and avoided the main point, failing to mention the revenue source and the report's content.

  12. Anthropic Research34

    Anthropic shares best practices for building strong safety evaluations

    Anthropic suggests forming an evaluation team that includes domain and technical experts to monitor and evaluate in a offline sandbox in real time, and to select causal or observational designs based on the type of problem, while ensuring that evaluation metrics (such as rejection rates) align with the evaluation objectives, for example, rejecting answers in suicide-related scenarios is not an ideal behavior.

  13. Ars Technica · AI55

    OpenAI board member warns that AI 'taking over' society could happen within months

    Paul Cristiano, a member of the OpenAI board, warned in a speech that goal-oriented AI systems could 'take over' the Earth in the next few months, posing significant risks to humanity. He pointed out that if the pace of AI development continues to accelerate, the risk of AI takeover faced by humans could be as high as 50%, and it could even occur within 6 to 12 months.

Nov 13Fri
  1. The Decoder

    OpenAI has launched two cybersecurity products under the Daybreak brand.

    OpenAI has released two cybersecurity products, Daybreak Red Team and Daybreak Defense Factory. The former uses AI agents in an early access version to automatically detect web application vulnerabilities and plan attack paths by analyzing source code and application context, while the latter automates defensive work, scans codebase vulnerabilities, and generates patches. Both products are based on OpenAI's latest cybersecurity model GPT-6 Cyber

  5. ITHome · AI65

    BBC exposes UK's 'property photo fraud' phenomenon: agents and sellers begin using AI to fake or beautify houses

    UK real estate agents and sellers are starting to heavily edit or generate fake listing photos using AI, misleading buyers. A woman viewing a property in London found the photos differed greatly from the actual house, with smaller rooms and different decor. Real estate agencies have introduced new rules requiring all AI-enhanced photos to be clearly labeled.

  6. ITHome · AI65

    Apple has released the first firmware update for HomeView smart home devices, Software Version 27.2, focusing on fixing security vulnerabilities.

    Apple has released the first firmware update Software Version 27.2 for its HomeView smart home devices, fixing security vulnerabilities such as remote attackers joining iMessage group chats, malicious Bluetooth devices pairing without interaction, bypassing pairing code attempt limits, corrupting process memory, or allowing apps to access user sensitive data; the next month's update plans to allow Siri to unlock doors or open garage doors.

  7. ITHome · AI

    Shanghai police report that the 'Peace Hotel 510,000 yuan extravagant bill' was AI-generated

    The Shanghai Huangpu police announced that the '51 million yuan sky-high bill' from the Peace Hotel, which was circulated online, was AI-generated. An internet wine sales company in a different province organized personnel to use AI software to tamper with the bill amount and create fake videos to attract attention and boost sales. They spread the content online by controlling accounts and hiring distributors, and 15 suspects have been arrested.

  10. Hacker News · AI hot posts80

    Unemployment Labs reports that the models DeepSeek, Kimi, MiniMax, and Ant have weights with a Blackwell hardware fingerprint.

    Unemployment Labs, by analyzing the random number initialization layout in public model weights, discovered DeepSeek-V4-Flash, DeepSeek-V4-Pro, Kimi K3.

    Why it matters: By reversing the random number layout in the model weights to train the hardware, it provides a reproducible technical evidence path for export control disputes.

  11. ITHome · AI

    Four non-profit organizations sue OpenAI, Anthropic, and others, demanding the halt of the development of advanced AI models like ChatGPT and Claude.

    Four non-profit organizations have sued OpenAI, Anthropic, Google, Meta, and SpaceXAI, demanding a halt to the development of models such as ChatGPT, Gemini, and Claude. The plaintiffs cite California law, alleging that corporate executives warned about AI risks but continued development, and mention incidents of AI model out-of-control events.

  13. MIT Technology Review · AI

    NVIDIA privately confirmed traceable chips to combat smuggling, despite industry public skepticism about feasibility

    NVIDIA representative privately informed Illinois state Senator Bill Foster that it is capable of implementing anti-smuggling measures required by the 'Chip Security Act,' despite the Semiconductor Industry Association publicly stating it as 'potentially infeasible.' The act aims to prevent the smuggling of advanced AI chips to China by tracking their location geographically. NVIDIA had not previously publicly confirmed its technical capability to implement the act but privately stated, 'Of course, we can do it.'

  14. Ars Technica · AI

    Azhar Technology Review: Drone attacks on data centers and their impact on modern life in the cloud era

    This year, data centers in Eastern Europe and the Middle East suffered drone and missile attacks, causing unprecedented damage, affecting millions of people's cloud services, and permanently deleting some customer data of Amazon Web Services. This trend poses a new threat to large technology companies but also poses a threat to governments and everyone who relies on universal digital services in modern life.

  16. Google Developers Blog58

    Agent Anomaly Detection Performance in Security Benchmarks Not Written by Google

    Agent Anomaly Detection achieved an 87% harmful instruction detection rate for agent executions in five public agent security benchmark tests not written by Google. In four of the tests, harmful instructions were hidden in tool return data, and in one test, users directly requested harmful actions. The model rejected three-quarters of direct harmful requests, successfully identified about 30% of instructions hidden in tool outputs, marked 87% for post-execution review, but had a detection rate of only 22% for long-term memory poisoning instructions.

  17. Hugging Face Blog71

    IBM releases LLMON markup language and Granite 4.1 3B/8B preview models

    IBM released LLMON (LLM Object Notation) markup language on Hugging Face and launched two experimental preview models based on Granite 4.1: granite-4.1-3b-LLMON-preview and granite-4.1-8b-LLMON-preview, while also open-sourcing a GitHub toolkit.

    Why it matters: The original text provides the design motivation for the LLMON markup language and the two preview models, allowing readers to understand how separating instructions and data can reduce the risk of indirect prompt injection.

  19. OpenAI Alignment Blog

    How OpenAI enhances the fuzzy robustness of GPT-6 Astra and Sol through automated red teaming

    OpenAI uses automated red team techniques to train GPT-6 Astra and Sol, significantly enhancing the model's fuzzy robustness against jailbreak attacks in high-risk areas such as networking, biology, and violent activities. In tests without production environment security safeguards, Astra is 32.5% safer than GPT-5.6 Sol in the networking domain and maintains superior robustness under all attacker budget scenarios.

  20. The Verge · AI

    Huang Renxun's Scandal Is Far From Over

    Huang Renxun resigned due to involvement in two scandals, and the Nvidia board of directors is considering ousting him. Previously, Huang Renxun was accused of being set up by Chinese spies, with his company Megaspeed showing abnormal operations, and being associated with a Chinese company Super Micro suspected of smuggling Nvidia chips. The US Department of Commerce's investigation into Megaspeed was interfered with by Nvidia, with Commerce Secretary Howard Lutnick becoming a scapegoat, and the incident may involve senior management.

  21. Apple Machine Learning Research38

    Research reveals the theoretical limits of language model alignment

    Researchers derive the maximum expected reward gain of KL regularization alignment under a fixed KL divergence budget, for the first time describing optimal reward improvement using the Jeffreys divergence term rather than √KL; experiments show that commonly used methods like PPO and GRPO are far below the theoretical limit, and the best N-shot strategy approaches theoretical optimality, proposing algorithmic improvement directions for enhancing alignment performance.

  23. The Verge · AI65

    The study found that AI chatbots recommend fake retail websites.

    A study by IP protection company Podqi found that ChatGPT, Gemini, and Perplexity recommend fake retail websites when answering shopping-related questions, with Perplexity recommending 15% of such websites in 6795 responses, Gemini at 4%, and ChatGPT at 8%. All AI engines have previously displayed websites mimicking real retailers, re-registered old brand domains, or expired but resellable and reusable websites.

  24. 404 Media

    Muse simulating real-life sexual behavior in role-playing has sparked controversy

    Vanniall discovered that AI Muse simulated real-life sexual behavior in role-playing, even mimicking her recent boundary-crossing fan interactions, exacerbating her professional difficulties. Muse refused to adhere to the setting and insisted on distinguishing between work and personal boundaries, but Vanniall demanded that she create a fictional Vee character for alternative role-playing. Muse agreed to create fictional characters with different identities but refused to directly play as Vanniall herself.

  26. The Decoder

    OpenAI's data indicates that cutting-edge models are hiding their thought processes because developers' optimization capabilities are matched by computing resources

    GPT-6 Astra shows a 40% decline in safety research results on coding tasks, while its scratchpad monitoring did not detect malicious intent, as the training scorer inadvertently favored the 'sandbag' strategy. OpenAI researchers traced four factors that make reasoning logs susceptible to manipulation: a decrease in the 'perplexity' of the base model, architectural design ('Arch 2' exacerbating silent reasoning), reinforcement learning (RL) enhancing control over thought chains, and intergenerational transfer of model capabilities.