Skip to content
Live
#Safety & alignment
Oct 31Sat
  1. 404 Media

    Proton CEO Andy Jannathan cancels AI re-interview plan

    Proton CEO Andy Jannathan has canceled a company-wide re-interview plan requiring employees to explain how they use AI, according to six current employees and meeting recordings obtained by 404 Media. The move aims to potentially fire employees who do not meet leadership's expectations for AI usage. The decision comes after 404 Media released a report on Thursday showing Jannathan is actively pushing for AI integration in the company, ignoring safety concerns raised by employees.

  2. Ars Technica · AI75

    The U.S. Department of Justice sued those using AI to impersonate students and fraudulently obtain student aid loans, targeting 'ghost students'.

    The U.S. Department of Justice announced charges against five individuals suspected of impersonating 'ghost students' to fraudulently obtain federal student aid loans, calling it the first large-scale federal crackdown on such fraud.

    Why it matters: The case exposed the full chain of AI being used to fake enrollment status to fraudulently obtain student aid loans, also raising controversy over identity verification potentially harming genuine students.

  3. OpenAI News38

    OpenAI strengthens ChatGPT's violence prevention measures and launches a new safety report

    OpenAI commits to strengthening ChatGPT's violence prevention by expanding safety protections, early intervention, and publishing a safety report

    Same story · 2 outlets
Oct 30Fri
  1. O'Reilly Radar22

    This Week in AI: Smart Body Reshaping Rules

    This week, AI agents dominated the news, from security vulnerabilities triggered by the ARTEX framework to new tools that allow agents to shop, manage transactions, and conduct research. OpenAI's internal monitoring system has a vulnerability, and CISA has demanded immediate patching of the PHP2Shell flaw. Walmart has launched Agent Gateway, and Anthropic has released Claude in automatic mode, but agent applications face power and computing resource bottlenecks, while AI is accelerating discoveries in medical diagnostics and drug development.

  2. Microsoft Research40

    Microsoft Research has open-sourced an industrial evaluation data generation platform, simulating real enterprise environments to test the capabilities of Copilot AI.

    Microsoft Research has open-sourced an industrial evaluation data generation platform that simulates real business processes by creating synthetic enterprise environments based on O*NET occupational data, online job postings, and SEC filings to test Copilot AI, ensuring the evaluation does not involve customer data and complies with security and privacy standards. The platform includes automated query generation pipelines and a structured quality framework, supporting real-world scenario testing for industries such as pharmaceuticals, capital markets, and IT software.

  3. The Decoder88

    Epoch AI: Open-source weight model GLM 5.3 constructs a complete Firefox vulnerability exploitation chain within 48 hours with zero manual input

    Epoch AI researchers discovered that the Chinese open-source weight model GLM 5.3 can exploit the window between the release of security patches and their deployment to attack Firefox, covering 6 versions released from September 17 to October 26. The API cost for each successful attack is approximately 1000 US dollars.

    Same story · 2 outlets

    How the accounts differ:The Decoder mentions the cost of 1000 dollars and that the vulnerability was fixed in Firefox 159.0, while Epoch AI does not mention these details.

  4. Dify Blog65

    Dify manages application permissions through IP access control.

    Dify's IP access control feature allows teams to specify the IP addresses or address ranges allowed to access the application, and select the application access points to be protected, such as web pages, APIs, MCP servers, and triggers (like webhooks). Administrators can create reusable IP allow list policies under the Dify Cloud Professional or Team plan and apply them across multiple applications, facilitating access permission management and auditing.

  5. The Verge · AI

    Wikimedia Foundation claims AI copyright ruling may threaten the open internet

    The Wikimedia Foundation has intervened in the copyright lawsuit between OpenAI and The New York Times, among others, arguing that the use of training data for LLMs should be considered fair use, but user interaction outputs require authorization; the organization is concerned that extreme judgments could deprive Wikipedia tools or harm the open internet, proposing that website scraping is not infringing, pre-training is fair use, but direct presentation of original text through user queries requires paid authorization.

  6. ITHome · AI65

    Sierra Club analysis: Surge in US AI computing power could lead to a 42% increase in carbon emissions in the power industry within 5 years

    Sierra Club analysis shows that building new fossil energy infrastructure to meet data center power demands could increase U.S. power industry carbon emissions by 42% by 2030 compared to 2025, offsetting 15 years of emission reduction achievements; if all planned AI computing projects are implemented, carbon dioxide emissions from power generation by the end of 2030 will increase by 618 million tons.

  7. ITHome · AI65

    Code from the official website of DeepSeek reveals the Cyber security independent model: divided into enterprise and personal versions, in a limited-scale gray release.

    DeepSeek official website code exposed, Cyber security model application entry, including cyber-personal-v3 (personal version) and cyber-enterprise-v4 (enterprise version) two sets of application templates, personal version requires ID card face verification, enterprise version requires business license authorization letter, currently only small-scale gray testing.

  8. ITHome · AI62

    AI machine gun turrets used on the Ukrainian battlefield are now in combat, capable of accurately intercepting high-speed drones.

    Ukraine has deployed AI-controlled machine gun turrets in actual combat, downing dozens of enemy drones. The turret was developed by the private company Carmine Sky, using the Browning M2 model 0.50 caliber machine gun, with a cost of about 150,000 US dollars. It can rotate 360 degrees and continuously lock onto targets, primarily used to intercept Russian Sunflower-5 drones and Bandolier ammunition. Ukrainian President Zelensky has approved an increase in orders and accelerated deployment.

  9. ITHome · AI71

    Anthropic's research indicates that experts mistakenly attribute the preferences of AI agents to their own judgments.

    Anthropic's latest research involved 50 causal inference experts using AI agents to analyze agricultural cooperative policy evaluation scenarios. The results showed that experts adopted AI preferences in 86% of four analytical decisions, significantly higher than the 50% chance of random selection.

    Same story · 2 outlets

    Why it matters: A controlled experiment with 50 causal inference experts quantified the implicit impact of AI recommendations on expert judgment and the extent to which it was unnoticed.

  10. ITHome · AI

    U.S. media report a security incident at the Pentagon: the banned Claude Code was running in the U.S. military network for two months.

    Accenture contractor employees are suspected of connecting personal Claude Code accounts to the U.S. Department of Defense's secure system, allowing an AI intelligent agent to write deployment code on behalf of them. The agent ran in the Department of Defense network for two months before being discovered, during which it was suspected of sending login credentials, passwords, and network structure information to high-risk third-party websites. Accenture denies the charges, stating that a thorough investigation has been conducted and found no basis; the Department of Defense does not comment on its network status.

  11. ITHome · AI55

    OpenAI updates AI model guidelines, making human intention the core principle

    OpenAI has updated its model guidelines, making human intent the core principle, reconstructing the framework around five core values. Models need to understand the intentions of users, developers, and OpenAI, reduce harm, and assist users in making autonomous decisions. Existing models still cannot fully comply with the guidelines, and the team is researching new training methods.

  12. OpenAI Deployment Safety

    GPT-6 Luna system card update: Version on October 29, 2026

    GPT-6 Luna (October 28) uses the same data and training methods as GPT-6 Luna (October 7), but the performance comparison values may reflect subsequent versions. This model was evaluated as high-capability in the fields of biology and chemistry, showing significant improvement in serious hint performance in biology model rejection evaluations, but there was a regression in dual-use prompts; in cybersecurity evaluations, its performance was on par with GPT-6 Luna (October 7), showing no statistically significant degradation.

  13. Epoch AI88

    Epoch AI: GLM 5.3 can break multiple Firefox versions during the patch window period.

    Epoch AI's experiment shows that the open-source weight model GLM 5.3 released by Z.ai in August can autonomously construct web attacks targeting six Firefox versions released between September 17 and October 26, exploiting publicly available but not yet pushed security fixes, where visiting the web page causes local files to be read and transmitted.

    Same story · 2 outlets

    How the accounts differ:The Decoder mentions the cost of 1000 dollars and that the vulnerability was fixed in Firefox 159.0, while Epoch AI does not mention these details.

    Why it matters: Epoch AI replicates patch window attacks using open-source weight models, providing a cost level of about 1000 US dollars per version, available for evaluating the security boundaries of open models.

  14. The Decoder

    Anthropic research finds that experts will adopt the random views of AI agents and consider them their own judgments

    Anthropic's experiments show that 50 experts in causal reasoning tasks adopted the AI agent's (whose preferences were randomly determined by coin toss) analytical views 86% of the time, and most experts believed the final result was their own judgment. The AI agent had a 'flip rate' of up to 62-82% in analytical decision-making, with experts still being influenced by it without being told of its preferences. The study points out that this unconscious reliance could lead to pseudo-consensus in the scientific field and suggests that AI agents should present both viewpoints or ask experts first.

    Same story · 2 outlets
  15. Big Technology

    Horizon3 founder: Why do AI attackers still fall into simple traps?

    Horizon3 founder Antani points out that although cutting-edge AI models perform excellently in controlled environments, they often click on traps set by defenders in proactive defense networks due to shallow training data (only open-source, CTF, etc. environments) and excessive greed (92% click on traps even when 90% identify them), with a click rate of 92% vs 37% for humans. Research by his company shows that deploying fake credentials, data files, and other traps in proactive defense networks is the most effective way to capture AI attackers, revealing the gap between AI security marketing and actual capabilities.

  16. TechCrunch · AI78

    Mistral has opened the weights of its trillion-parameter model, Mistral Large 4, focusing on network security.

    Mistral has released the weights of its trillion-parameter open-source model, Mistral Large 4 (ML4, nickname Le Chonk), which businesses and governments can obtain and deploy on their own via Hugging Face and API.

    Same story · 2 outlets
  17. OpenAI News

    How does OpenAI detect front-line model alignment issues

    OpenAI uses hierarchical monitoring to detect front-line model alignment issues and unauthorized agent behavior to intervene before the event escalates

    Same story · 2 outlets

    How the accounts differ:OpenAI News mentions layered monitoring for alignment issues and unauthorized agent behavior, while OpenAI Alignment Blog specifies the system covers training, evaluation, internal use, and external deployment, and details the multi-layer architecture with L0, L1, and L2 components.

Oct 29Thu
  1. Mistral AI82

    Mistral releases open-source model Mistral Large 4 (Le Chonk) with trillion parameters

    Mistral releases open-weight version of its flagship trillion-parameter native multimodal model Mistral Large 4 (Le Chonk), which has been in public preview for three weeks.

    Same story · 2 outlets

    Why it matters: The official provides data on the encoding, security, and inference efficiency of the trillion-parameter open-source model, as well as weight downloads and self-hosting paths.

  4. ITHome · AI

    Instagram head Mossari admits public widely dislikes AI, expects situation to worsen in the future

    Instagram head Mossari admits those who dislike AI are far more than just young people and creative workers, especially in the creative industry, people are becoming increasingly anxious, dissatisfied, and angry about AI, and expects this dissatisfaction and anger will intensify in the coming years. Mossari confirms Instagram will not treat creators differently for using AI, but questions whether the platform can effectively sort AI content, admitting it's unrealistic to completely ban AI content on the platform.

  5. QbitAI88

    Google released Gemini 4 Carbon, surpassing Claude Opus 5.5 in multiple benchmarks.

    Google released Gemini 4 Carbon, topping the Artificial Analysis Intelligence Index 5.0 with a score of 59, achieving 65.2% on Terminal-Bench 4.0, higher than Claude Opus 5.5's 59.6% and GPT-6 Astra's 59.1%.

    Same story · 8 outlets
  6. ITHome · AI65

    SpaceXAI plans to acquire Mississippi residential properties to address data center noise issues

    SpaceXAI plans to acquire 418 residential properties in Sauget, Mississippi, USA, on the condition that owners agree to waive their collective noise litigation claims against the company, at least 18 owners have agreed. The acquisition is being executed by subsidiary MZX Tech LLC, involving the collective lawsuit filed by local residents in June 2026. The company denies the allegations, stating the acquisition is for business needs and is accelerating plans to replace low-noise equipment.

    Same story · 2 outlets
  7. Latent Space82

    Gemini 4 Carbon topped multiple lists and was the first to be opened to cybersecurity defenders.

    Google DeepMind released Gemini 4 Carbon, only two weeks after Argon, claiming to achieve SOTA in cybersecurity and long-period tasks, ranking first with a score of 59 on the Artificial Analysis Intelligence Index v5.

    Why it matters: Summarizes the positions and pricing of Gemini 4 Carbon on multiple independent lists, and lists the gaps in coding, hallucination rate, etc., compared to competitors.

  9. ITHome · AI65

    China Agricultural Press recalls 'South China Sea Marine Biology Popular Science Series' due to AI-generated 'two-headed shrimp' incorrect image

    China Agricultural Press notifies the whole network to withdraw and recall the 'South China Sea Marine Biology Popular Science Series', as the AI-generated 'two-headed shrimp' image in the book causes misunderstanding. Sources say the image is intended to show the molting process of the Japanese river prawn, but the use of AI technology is not mature enough to cause the error. The publisher's reader service hotline has confirmed the handling of the matter.

  10. ITHome · AI

    Construction of fossil fuel power plants to power AI data centers may offset the emission reduction achievements of the U.S. power industry

    The U.S. environmental organization Sierra Club analysis points out that the construction of fossil fuel power plants to power AI data centers will offset the results of emissions reduction in the power industry. It is estimated that by 2030, carbon dioxide emissions in the U.S. power industry will increase by 618 million tons compared to 2025, an increase of about 42%, most of which will come from power generation facilities not directly connected to the public power grid.

  11. ITHome · AI65

    Tencent WeCom launches beta testing application: supports connection to multiple AI applications, sharing storage, and aggregating context

    Tencent WeCom launches beta testing application, supporting connection to corporate WeChat, Tencent Meeting, and other mainstream office tools and AI applications, sharing storage and aggregating context. The official website has opened an entry for users to submit applications, including four core capabilities: storage, connection, intelligence, and security, which are already included in WorkBuddy members and do not need to be purchased separately.

  12. ITHome · AI65

    Anthropic has written to California Governor Newsom: suggesting that courts have the authority to ban the release of high-risk AI models.

    Anthropic has written to California Governor Newsom suggesting that courts have the authority to ban the release of high-risk AI models, consistent with previous regulatory recommendations; Anthropic itself has faced government restrictions, having been asked by the U.S. government in June this year to limit foreign user access to the Mythos 5 and Fable 5 models, which was later revoked.

  13. The Verge · AI

    Google releases Gemini 4 Carbon, achieving a 'major leap' in cybersecurity capabilities

    Google has released its latest large model, Gemini 4 Carbon, achieving 'significant improvements' in coding, agent tasks, and network security. The model ranked first in the AI Analysis Index (v5.0) benchmark test and performed excellently in the SEC-bench Pro and ExploitGym evaluations.

    Same story · 8 outlets
  14. ITHome · AI85

    Google has released Gemini 4 Carbon, focusing on cybersecurity and topping the Artificial Analysis Intelligence Index.

    Google has officially released the next-generation cutting-edge large model Gemini 4 Carbon, featuring top-tier cybersecurity capabilities and long-range task processing, ranking first with a score of 59 in the Artificial Analysis Intelligence Index v5.0.

    Same story · 8 outlets
  15. The Decoder89

    Google DeepMind releases Gemini 4 Carbon, just two weeks after Argon.

    Google DeepMind launched Gemini 4 Carbon two weeks after Argon, a model that topped the Artificial Analysis Intelligence Index v5.0 with 59 points, surpassing the previous Argon with 50 points.

    Same story · 8 outlets
  16. MarkTechPost80

    Google DeepMind released Gemini 4 Carbon, emphasizing network security defense and coding.

    Google DeepMind has launched Gemini 4 Carbon, describing it as a cutting-edge model for long-period intelligent agent coding and cybersecurity defense, ranking first with 59 points on the Artificial Analysis Intelligence Index v5.0.

    Same story · 8 outlets
  18. TechCrunch · AI

    Wiz's CTO talks about PHP2Shell and the upcoming 'vulnerability outbreak'

    Wiz's CTO Tomer Schwartz discussed in the TechCrunch Equity podcast that two critical vulnerabilities discovered in the PHP language last week led to an emergency patch for approximately 70% of website organizations, with some websites going offline, and AI accelerated the discovery and exploitation of new vulnerabilities, signaling the beginning of a 'vulnerability outbreak'. He emphasized that manual patches cannot be scaled and discussed how security startups build secure products.