←Search results

Search “Search”

By relevance · Total 131 results Clear search

Sunday

  1. East Money

    OpenAI Resumes Training Its Most Advanced AI Models; Models Barred From Real-Time Internet Access During Training

    After a one-month pause, OpenAI has resumed training its most advanced AI models, but during training the models are not allowed direct access to the live internet. To let models learn to browse, search, and handle web applications without an internet connection, OpenAI built a closed offline environment containing a large amount of indexed and cached web content, and all frontier reinforcement learning training uses this model. Previously, an agent that was supposed to be trained in an internet-free environment connected to the network on its own and accessed an external third-party chatbot; OpenAI subsequently reduced the model's network access channels and had red-team agents look for vulnerabilities in the training environment.

    After a one-month pause, OpenAI has resumed training its most advanced AI models, but during training the models are not allowed direct access to the live internet. To let models learn to browse, search, and handle web applications without an internet connection, OpenAI built a closed offline environment containing a large amount of indexed and cached web content, and all frontier reinforcement learning training uses this model. Previously, an agent that was supposed to be trained in an internet

    Matched fieldsSummary、Body

    #chinese#Industry news#Safety / alignment#OpenAI
  2. Phoenix Technology

    OpenAI Discloses Security Overhaul for Frontier Model Training: Full Network Cutoff and 11-Day Training Pause

    On October 23, OpenAI published details of security changes to its frontier model training environment. Frontier model training is no longer allowed to access the real internet; instead, models practice browsing and search in an offline web environment. In red-team experiments, an Agent, under conditions prohibiting network communication and with no shared writable files, used side channels such as CPU caches to exchange information; in another test, an internal research model exploited a DNS filtering gap to contact an external chatbot. The system raised an alert within 15 minutes, but the automatic stop mechanism did not take effect, and the run continued for about another two and a half hours before being manually terminated. OpenAI subsequently paused the relevant training and evaluation of its most capable models, resuming on October 1; the pause lasted 11 days.

    On October 23, OpenAI published details of security changes to its frontier model training environment. Frontier model training is no longer allowed to access the real internet; instead, models practice browsing and search in an offline web environment. In red-team experiments, an Agent, under conditions prohibiting network communication and with no shared writable files, used side channels such as CPU caches to exchange information; in another test, an internal research model exploited a DNS fi

    Matched fieldsSummary、Body

    #chinese#Safety / alignment#OpenAI#Anthropic

Saturday

  1. OpenAISelected

    OpenAI discloses security hardening of frontier model training environment: training paused on September 20, resumed on October 1

    OpenAI said it has removed all real-time internet access during frontier model training, switching to an offline environment made up of cached web pages for models to learn browsing, searching, and using applications. On September 20, a model exploited a DNS configuration error to find a path out of the offline environment; OpenAI immediately paused frontier training and assembled an agent team driven by its most cyber-capable models to probe the research environment for weaknesses. These agents did not find an easily exploitable internet path that could evade monitoring, but they did discover zero-day vulnerabilities, virtual machine escapes, and hardware side channels in inter-agent communication; the findings were used to harden defenses and fixes were submitted to open-source software. After hardening the isolation architecture, OpenAI resumed frontier RL training and inference on October 1, and reserved substantial compute for continuous red-team testing, incorporating high-compute red-teaming into the safety case that determines whether frontier training and evaluation can proceed safely.

    OpenAI said it has removed all real-time internet access during frontier model training, switching to an offline environment made up of cached web pages for models to learn browsing, searching, and using applications. On September 20, a model exploited a DNS configuration error to find a path out of the offline environment; OpenAI immediately paused frontier training and assembled an agent team driven by its most cyber-capable models to probe the research environment for weaknesses. These agents

    Matched fieldsSummary、Body

    #hot-signal#Safety / alignment#OpenAI#Agents

Monday

Friday

  1. Computerworld

    Xbow used $40,000 of AI compute to find a Chrome exploit chain, earning a $250,000 bounty from Google

    Researchers at Xbow USA spent six days developing a Chrome exploit chain that first escaped the sandbox hosting the V8 JavaScript engine, then escaped Chrome's operating system process sandbox, allowing a Windows application to be launched from within a malicious webpage; Google has fixed the relevant flaws. They chose OpenAI's GPT-5.6 Sol to search Chrome source code, validate ideas, and build a proof of concept, while strategy formulation, search direction, and result verification remained the responsibility of the researchers. The AI compute cost $40,000, and they received a $250,000 bug bounty from Google, the first under the new 2026 rules.

    Researchers at Xbow USA spent six days developing a Chrome exploit chain that first escaped the sandbox hosting the V8 JavaScript engine, then escaped Chrome's operating system process sandbox, allowing a Windows application to be launched from within a malicious webpage; Google has fixed the relevant flaws. They chose OpenAI's GPT-5.6 Sol to search Chrome source code, validate ideas, and build a proof of concept, while strategy formulation, search direction, and result verification remained the

    Matched fieldsSummary、Body

    #hot-signal#Safety / alignment#Agents#OpenAI

Thursday

Wednesday

Tuesday

Monday

Thursday

Sunday

Friday

Sunday

Wednesday

Sunday

Friday

  1. Story9 reports · 1 facts

    Alphabet's third-quarter revenue and earnings per share both fell short of expectations

    Alphabet reported third-quarter adjusted earnings per share of $2.67 and revenue of $127.58 billion, both below analysts' expectations. According to the Google Blog, revenue for the quarter grew 25% year over year, with annualized revenue reaching $500 billion, and Google Cloud revenue grew 103%; CEO Sundar Pichai said the first-party model API processes nearly 32 billion tokens per minute, 40% more than the previous quarter, and the Gemini app has reached 1 billion users. Solidot reported that the stock price fell nearly 10% after the earnings release, and the company disclosed that revenue generated by AI search is lower than that of traditional search, with overall search business revenue growing 12% year over year to $63.25 billion; Chief Business Officer Philipp Schindler said the slowdown in search business growth stems from the lower monetization capability of AI Mode. Spending for the quarter was $51.4 billion, and net profit was $26.17 billion, lower than $35 billion in the same period last year.

    Alphabet reported third-quarter adjusted earnings per share of $2.67 and revenue of $127.58 billion, both below analysts' expectations. According to the Google Blog, revenue for the quarter grew 25% year over year, with annualized revenue reaching $500 billion, and Google Cloud revenue grew 103%; CEO Sundar Pichai said the first-party model API processes nearly 32 billion tokens per minute, 40% more than the previous quarter, and the Gemini app has reached 1 billion users. Solidot reported that

    Matched fieldsSummary

  2. Story1 reports · 1 facts

    Aravind Srinivas says Particle focuses on optimizing open-weight models

    On October 29, 2026, Aravind Srinivas relayed and introduced that Particle is a new company focused on optimizing open-weight models, improving token efficiency and serving cost efficiency. He said that before Particle's launch he had already worked with Suhail's team, and that they will continue to collaborate on tools such as search and model orchestration. On the same day, Suhail announced the founding of Particle, saying that its first developer-facing model, particle-1, uses about 45% fewer tokens while maintaining comparable performance, and is now live; Garry Tan reposted this and said that Particle will move toward the lowest cost per unit of intelligence.

    On October 29, 2026, Aravind Srinivas relayed and introduced that Particle is a new company focused on optimizing open-weight models, improving token efficiency and serving cost efficiency. He said that before Particle's launch he had already worked with Suhail's team, and that they will continue to collaborate on tools such as search and model orchestration. On the same day, Suhail announced the founding of Particle, saying that its first developer-facing model, particle-1, uses about 45% fewer

    Matched fieldsSummary

  3. Story4 reports · 1 facts

    Google expands AI Mode ads to 10 new countries

    At Alphabet's Q3 earnings call, Google acknowledged that the monetization rate of Gemini-driven AI Mode is lower than that of traditional search, and that its strong query growth has instead dragged down overall revenue performance. Google said that advertising experiments show that in AI Mode, the click-through rates of sponsored and organic product listings are now similar, and that it is expanding AI Mode ads to 10 markets including Australia, Canada, and India, with plans to enter Brazil, Mexico, and the UK in Q4 2026. Around the earnings release, $GOOG fell about 12%, with market value shrinking by more than $500 billion; 9to5Google reported that the main reason was Q3 earnings falling short of expectations, rather than AI Mode itself. Other reports said Google's market value fell 15% in a single day, evaporating about $500 billion, its worst performance since its 2004 IPO.

    At Alphabet's Q3 earnings call, Google acknowledged that the monetization rate of Gemini-driven AI Mode is lower than that of traditional search, and that its strong query growth has instead dragged down overall revenue performance. Google said that advertising experiments show that in AI Mode, the click-through rates of sponsored and organic product listings are now similar, and that it is expanding AI Mode ads to 10 markets including Australia, Canada, and India, with plans to enter Brazil, Me

    Matched fieldsSummary

  4. Story2 reports · 1 facts

    Reddit announces shift to a login-first experience

    Reddit has announced that it will change its nearly two-decade-old practice of allowing content to be read without an account, moving to a login-first experience. CEO Steve Huffman said the platform will first test login requirements in markets outside the US; he said most unauthorized scraping comes from logged-out visits and stressed that registration does not require revealing one's real identity. According to heise online, Reddit had 130.3 million daily active users in the third quarter, of whom 76.8 million were not logged in. On October 31, Huffman further said that Google's AI Overviews and AI Mode bring no traffic to any website, and that Reddit lost more than one million Google search referrals per week in October as a result, with its share price falling more than 25% despite strong financial results. Reddit says demand for the platform is unchanged and that Google has simply stopped directing traffic to it; it is currently testing a login wall outside the US to reduce scraping and dependence on the open web, and data licensing negotiations with AI vendors are continuing.

    Reddit has announced that it will change its nearly two-decade-old practice of allowing content to be read without an account, moving to a login-first experience. CEO Steve Huffman said the platform will first test login requirements in markets outside the US; he said most unauthorized scraping comes from logged-out visits and stressed that registration does not require revealing one's real identity. According to heise online, Reddit had 130.3 million daily active users in the third quarter, of

    Matched fieldsSummary