The week's main developments, grouped by original publication date. Every item links to our full story and the original source.
Anthropic reports that the Claude model produced a computer-checked proof of Fermat’s Last Theorem in 11 days, generating 13 million lines of Lean code.
This project highlights the potential for AI-assisted formalization to rigorously verify complex mathematical proofs, potentially streamlining the validation process for new research.
OpenAI has introduced GPT-6 Astra, a model achieving top marks on complex reasoning benchmarks and improved autonomous computer operation.
This release marks a shift toward highly autonomous technical workflows, highlighting the dual-use nature of models capable of both vulnerability discovery and defense.
Google has introduced WeatherNext 3, an AI model that generates hourly high-resolution weather forecasts using live satellite data.
This update provides industries such as renewable energy and agriculture with granular, real-time climate insights. By utilizing live satellite mosaics instead of traditional numerical prediction inputs, Google aims to improve reliability across its ecosystem.
Google has launched Gemini 3.8 Flash for agentic reasoning and a specialized Cyber version for automated vulnerability detection and patching.
These models provide enterprises and developers with cost-efficient tools for autonomous agent workflows and automated software security patching.
Gemini 3.7, 3.6, and 3.5 Flash models now support agentic video analysis, reducing token costs and improving accuracy for long-form content.
This update enables developers to process long-form video more efficiently. By focusing on relevant segments rather than total frame counts, it optimizes resource usage for tasks like anomaly detection, object tracking, and sub-second retrieval.
Anthropic introduced Claude Fable 5.1, a high-capacity model designed for multi-stage workflows, featuring tiered safety protocols and vision capabilities.
This release provides enterprise users with a specialized tool for autonomous, multi-stage research. The tiered safety routing aims to maintain data integrity in sensitive domains.
OpenAI commits $1 billion to provide cybersecurity AI tools and support for organizations managing essential public services.
The initiative addresses the need for robust cybersecurity in essential infrastructure, such as utilities and local governments. By expanding access to advanced AI security tools, it enables organizations to strengthen their defenses against increasingly sophisticated digital risks.
CrowdStrike introduces SafeMind, an agentic security system leveraging NVIDIA Nemotron models for enhanced automated threat detection and response.
This move signifies a shift toward autonomous, agent-based security architectures that can evolve alongside adversarial tactics, potentially improving response speeds and operational efficiency for security teams.
Basis, Clay, and Exa Labs utilize AI agents for onboarding, account management, and developer tasks with mandatory human-in-the-loop review points.
The examples suggest a practical pattern: define a repeatable task, give the agent relevant context, and keep a person responsible for decisions. The reported time savings come from the companies involved.
Salesforce has introduced MCP server support to allow AI agents to securely access enterprise data and business logic without custom integrations.
By exposing business logic and permissions through open MCP standards, Salesforce allows organizations to connect AI agents to enterprise workflows without needing to build and maintain complex, custom integration layers.
NVIDIA Personal AI Router enables users to cluster local compute resources to speed up multi-agent AI workflows without infrastructure overhauls.
PAIR allows users to scale their AI inference capabilities by aggregating resources across a network, effectively reducing latency and queue times. This enables more complex AI applications to run on hardware that would otherwise be insufficient for the workload.