OpenAI has revealed Astra, an unreleased model designed to tackle complex, long-running tasks, after an internal version produced ten significant advances in mathematics and theoretical computer science. [...]

OpenAI has announced an unreleased artificial intelligence model named Astra, which the company states has achieved significant breakthroughs in mathematics and theoretical computer science. An internal version of Astra reportedly solved ten long-standing problems in these fields, some of which had seen no progress on their central results for over a decade.
The research conducted by OpenAI with Astra spanned diverse areas, including high-dimensional geometry, coding theory, arithmetic circuit complexity, group theory, quantum complexity, lattice cryptography, and extremal combinatorics. Specific examples of problems Astra is credited with solving include demonstrating the existence of non-sofic groups, disproving Connes' rigidity conjecture, establishing new bounds for high-dimensional sphere packing, and resolving several problems originally posed by mathematician Paul Erdős.
OpenAI indicated that the computational resources required for Astra to find solutions to these problems would equate to approximately $2,000 if priced at current Sol API rates. Following the AI's initial problem-solving, human researchers utilized the same model to formulate the arguments into manuscript form. Astra then formalized each argument as a Lean certificate, enabling verification through a mathematical proof-checking system.
Industry sources have independently corroborated that OpenAI is developing Astra as a new family of models specifically designed for managing long-running computational tasks. OpenAI describes Astra as a robust model that facilitates collaboration among AI agents on various components of a larger problem.
While the company has confirmed Astra's existence and capabilities, OpenAI has not yet finalized a release name for the model. Potential designations include GPT-5.7, GPT-6, or an entirely different name. Given its significant capabilities, there is speculation that Astra could be subject to a tiered release strategy, similar to policies seen with other advanced AI models, where a consumer-facing version is released alongside a more powerful variant requiring special approval.
A weakness has been identified in Tenda CP3 27.5.57.101. This issue affects some unknown processing of the file Net/NetCheckPing.cpp. This manipulation of the argument interface_name/host causes os command injection. The attack can be initiated remotely.
A security flaw has been discovered in Tenda CP3 27.5.57.101. This vulnerability affects the function SystemAsh of the file Apis/system.c of the component Kylin. The manipulation of the argument AlarmVoiceURL results in os command injection. It is possible to launch the attack remotely.

OpenAI has announced a $1 billion commitment to provide subsidized access to its Daybreak AI cybersecurity tools for under-resourced critical infrastructure defenders. The initiative, named Daybreak for Frontline Defenders, will offer AI models, training, and technical support over the next six months, prioritizing water and wastewater utilities, electric grid operators, and local government entities. This move aims to equip organizations with limited budgets and staff against increasingly sophisticated cyber threats.

Attackers are exploiting a new unpatched vulnerability in Magento Open Source and Adobe Commerce that lets them run malicious code on an online store's server without logging in, Dutch e-commerce security company Sansec said in an advisory published on September 5. Sansec, which discovered the flaw and named it StyleSmuggler, said attacks started on September 4. "Sansec is publishing early
In BPF instructions that load/store a value from/to a scratch memory register the register index is an unsigned 32-bit integer and must not exceed 15, but libpcap BPF interpreter does not validate the value. In particular uncommon use cases a crafted filter program can cause the interpreter to try reading and writing the OS process memory in the 16GiB starting at the current stack frame on 64-bit architectures and in the entire address space on 32-bit architectures.

Attackers are exploiting two new PaperCut flaws to steal credentials and gain privileged access in education-sector attacks across the U.S. and Europe. Attackers are exploiting two recelty disclosed PaperCut flaws, CVE-2026-81578 and CVE-2026-82078, in attacks targeting schools and other education organizations in the U.S. and Europe, as reported by TheHackerNews. Arctic Wolf researchers observed