#искусственный-интеллект
-
OpenAI Reworks Codex Into a Tool for Autonomous Workflows
OpenAI has unveiled an updated version of Codex capable of carrying out tasks on its own over the course of several hours. The company positions the tool as a way to work “for the user and with them.”
-
OpenAI Shuts Down Its ChatGPT Atlas Browser Less Than a Year After Launch
OpenAI is winding down its ChatGPT Atlas browser project, which could carry out tasks on the user's behalf. The decision was announced as part of the ChatGPT Work news.
-
Microsoft Will Ship More Security Fixes Thanks to AI
Microsoft says that using artificial intelligence to hunt for vulnerabilities will lead to a growing number of fixes in Windows 11's monthly updates.
-
Fidji Simo Steps Down as Head of OpenAI's AGI Division Due to Illness
Fidji Simo is moving from a full-time role to that of a part-time advisor at OpenAI. The reason is a neuroimmune condition that had previously already prompted her to take medical leave.
-
The New York Times Accuses OpenAI of Concealing Evidence in Copyright Case
Publishers claim that OpenAI hid tools and datasets capable of confirming that their material was used in ChatGPT's responses. The publications have filed a motion for sanctions, escalating their legal dispute with the company.
-
OpenAI Releases GPT-5.6 for Everyone After U.S. Government Approval
After two weeks of restricted access limited to government organizations, OpenAI has received approval from the Trump administration for a public launch of GPT-5.6. At the same time, the company unveiled a new product called “ChatGPT Work.”
-
Ollama raises $65M as it grows to nearly 9 million users
Ollama, the open-source tool for running AI models locally, has secured $65M in funding backed by Benchmark. The project has amassed 176,000 stars and nearly 17,000 forks on GitHub.
-
How the Language of AI Could Change Human Speech
Bruce Schneier warns: large language models are trained only on written texts and scripted speech, and the mass use of AI risks changing the way people themselves talk.
-
New Forg365 Phishing Platform Uses AI to Steal Microsoft 365 Accounts
Researchers have uncovered Forg365, a phishing platform that uses AI to generate lures for stealing Microsoft 365 credentials. The service runs on a phishing-as-a-service model and combines several defense-evasion techniques.
-
Meta is developing smart glasses that continuously record audio and photos
According to the Financial Times, Meta is working on a prototype of “super sensing” smart glasses capable of continuously recording sound and taking snapshots every few seconds. Users will be able to ask Meta AI for information about the captured material.
-
OpenAI Releases New Voice Models for Live Conversation
The company has unveiled an upgraded voice mode capable of speaking and listening at the same time. That capability is especially important for real-time simultaneous translation.
-
xAI Releases Grok 4.5 — an Opus-Class Model, According to Musk
Elon Musk's company has unveiled a new version of its Grok model, calling it a cheaper and more efficient alternative to other powerful AI systems.
-
CISA Orders U.S. Federal Agencies to Urgently Patch a Vulnerability in Langflow
The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has directed federal agencies to remediate an actively exploited authentication bypass vulnerability in the Langflow platform by Friday.
-
DeepSeek Plans to Develop Its Own Chips to Sidestep US Export Controls
The Chinese company DeepSeek plans to build its own chips in order to reduce its dependence on Nvidia and Huawei. The project is in its early stages.
-
Microsoft Cuts AI Spending, Betting on Its Own Models
Microsoft has joined the ranks of tech giants scaling back their artificial intelligence spending. The company plans to rely more heavily on its own models.
-
GitLost: researchers tricked GitHub's AI agent into exposing private repositories
Researchers demonstrated an attack called GitLost that deceived GitHub's AI agent into leaking data from private repositories. The write-up on the research drew notable attention on Hacker News.
-
Google Sues Chinese Scammers Who Used Gemini for Phishing
Google has accused a group called Outsider Enterprise of running phishing-as-a-service through Telegram using the Gemini AI model. The scammers built fake sites impersonating Google, YouTube and government agencies.
-
Tencent Releases Open MoE Model Hy3 with 295 Billion Parameters
Tencent has unveiled the Hy3 language model under the Apache 2.0 license, which, according to its developers, outperforms models in its class and comes close to flagship open-source solutions with substantially larger parameter counts.
-
The Challenge of Identifying AI Agents in Security
The blog's author raises the topic of securing the identity of autonomous AI agents. The piece was published on codon.org.uk.
-
Anthropic explores a “global workspace” in language models
Anthropic has published research on the concept of a global workspace as applied to language models. The piece drew notable attention on Hacker News.
- More