#ии
-
Meta Releases Muse Code — an AI Agent for Large Repositories
Mark Zuckerberg's company has unveiled a terminal-based coding agent, promising it can handle complex tasks in large codebases. The product has launched in beta.
-
Jeff Dean Is Leaving Google to Speed Up Science with AI
One of Google's longest-serving executives, together with several of the company's leading researchers, has founded Discovery Loop, a startup that aims to automate scientific experiments.
-
Meta Ran Ads Featuring AI-Generated Child Sexual Abuse Material
Researchers uncovered dozens of paid ads in Meta's ad library containing AI-generated images of child sexual abuse — all of which passed the company's moderation.
-
Three Paperclip Vulnerabilities Allow Command Execution via Malicious Agent Import
Three flaws — including a critical one with a maximum CVSS score of 10.0 — have been found in Paperclip, the open-source platform for managing teams of AI agents. All of them boil down to the same thing: an agent's configuration turns into executable code.
-
Rust Adopts LLM Usage Rules for rust-lang/rust Contributors
Five teams within the Rust project have adopted a policy governing the use of large language models when working on the main repository. The core principle: AI may be used to analyze and verify, but not to create code.
-
CISA Adds Actively Exploited Flaws in Langflow, Tomcat and N-central to Its Catalog
On August 5, 2026, CISA added three vulnerabilities to its KEV catalog, citing evidence that they are being exploited in real-world attacks. One of the flaws is tied to an autonomous AI-driven hacking campaign.
-
OpenAI Calls Apple's Trade-Secret Theft Suit ‘Aggressive and Oddly Personal’
OpenAI has publicly responded to a lawsuit in which Apple accused the startup of stealing confidential technology. The company says it holds — and wants — none of anyone else's trade secrets.
-
Public Claude Chats Turned Up in Google Search Results
Conversations with Claude containing private crypto-wallet keys and personal data ended up in Google's index. Anthropic's response: users have only themselves to blame.
-
AI Content Labeling Rules Take Effect in the EU
As of August 2, companies are required to tell users when they are talking to artificial intelligence and to label AI-generated deepfakes. Violations carry fines of up to €15 million.
-
AI-“Journalist” News Site Pushes the Interests of OpenAI's Super PAC
The outlet Acutus passes AI-generated text off as independent journalism. An investigation ties it to a political machine funded by OpenAI's inner circle.
-
Qwen3.8-Max: a new bar for programming and collaboration
The Qwen team has unveiled the Qwen3.8-Max model, positioning it as a tool for writing code and for collaborative work. The announcement drew notable attention on Hacker News.
-
Pippa Pays Artists for Their Style — but Is That Enough to Reconcile Them with AI?
The startup Pippa offers illustrators royalties for every image and video it generates, promising “ethical” generative AI. Yet its models are still trained on data scraped from the web without the creators’ consent.
-
Kimi K3 on AMD MI355X: Better Cost per Token than NVIDIA B300
Wafer's engineers ran the open 2.8-trillion-parameter Kimi K3 model on AMD MI355X accelerators and got noticeably better performance per dollar than on NVIDIA's flagship Blackwell.
-
ByteDance unveils Seedance 2.5: 30-second clips in a single pass
The ByteDance Seed team has released a new version of its generative video model, Seedance 2.5 — with 30-second clips in a single pass, an expanded set of references and frame-by-frame editing.
-
A False “Refutation” of the Collatz Conjecture Exposed a Bug in Lean's Kernel
A soundness bug has been found in the Lean proof system's kernel: with the help of AI, a formally accepted but incorrect “refutation” of the Collatz conjecture was constructed. Lean's author, Leonardo de Moura, has published a breakdown of what happened.
-
Reddit Questions the Value of Google AI Overviews Amid Falling Stock
Reddit CEO Steve Huffman has cast doubt on the value of the AI Overviews feature, and the company is reportedly considering breaking off its licensing deal with Google.
-
GNOME Introduces an RFC Process and Battles a Flood of AI Extensions
The GNOME project closed out July working on a formal RFC process to govern its development while simultaneously fighting off a surge of sloppy, neural-network-generated GNOME Shell extensions.
-
Startup Offered a Tattoo in Exchange for an Interview — and Apologized
AI startup LemonLime lured “reckless” developers with free tattoos of its logo. After a wave of criticism, the company's head admitted he had “gone too far.”
-
Opus 5 Is More Resistant to Prompt Injection
Anthropic's new model posted the best result on the IPI benchmark, outperforming both previous versions of Claude and all of its competitors. But it's still impossible to fully defend against prompt injections.
-
84 Vulnerabilities in 4G and 5G Network Cores: From Denial of Service to Session Hijacking
Researchers from Nanyang Technological University have uncovered an entire class of vulnerabilities in open-source implementations of mobile cores, born of the “blind trust” between network functions.
- More